# Get Agent Loadout — full data (last checked 2026-10-21)
Prices are taken from official pricing pages (or a named tracker) on the last_checked date. Anything we couldn't confirm is marked unverified.
Rankings are never for sale.
Some links may be affiliate links; it never affects rankings. Sponsored items are labeled "Sponsored". Sponsorship: https://getagentloadout.com/sponsor/

## Loadout: Ship a vibe-coded SaaS MVP
URL: https://getagentloadout.com/loadouts/saas-mvp/
Job: Go from idea to a paid, deployed web app with auth, a database and checkout.
Category: Vibe coding | Difficulty: Intermediate
Tools: Claude Code (https://claude.com/product/claude-code); Supabase (https://supabase.com); Vercel (https://vercel.com); Stripe (https://stripe.com)
Model: Claude Sonnet 5.5 (default) / Opus 5.5 for planning
MCP servers: Supabase MCP (https://mcp.supabase.com/mcp); Vercel MCP (https://mcp.vercel.com); Stripe MCP (https://mcp.stripe.com)
Estimated monthly cost: $20–$70/mo. Claude Pro $20. Free tiers of Supabase and Vercel work while you build; once you charge money plan for Supabase Pro $25 and Vercel Pro $20 (Hobby is non-commercial). Stripe takes 2.9% + 30¢ per US card charge.
Config (.mcp.json):
{
  "mcpServers": {
    "supabase": { "type": "http", "url": "https://mcp.supabase.com/mcp?project_ref=YOUR_PROJECT_REF" },
    "vercel":   { "type": "http", "url": "https://mcp.vercel.com" },
    "stripe":   { "type": "http", "url": "https://mcp.stripe.com" }
  }
}
Starter prompt: Read PLAN.md. Build the smallest version of this SaaS that a stranger could pay for: email login (Supabase Auth), one core feature, a Stripe Checkout page for a single monthly plan, and a settings page. Use Row Level Security on every table. Before writing code, list the tables and RLS policies and wait for my OK. After each step, run the app and tell me exactly what to click to test it.
Steps:
1. Write a one-page PLAN.md: who it's for, the one feature, the price.
2. Create a Supabase project and a Vercel project; add the three MCP servers and authenticate each with /mcp.
3. Run the starter prompt in Claude Code and approve the schema first.
4. Deploy a preview on Vercel, test checkout in Stripe test mode, then switch to live keys.
Watch out: Point the Supabase MCP at a development project, not production. Add read_only=true to the URL when you only need it to inspect data.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Landing page in an hour
URL: https://getagentloadout.com/loadouts/landing-page-hour/
Job: Launch a good-looking landing page with a waitlist, today, without touching a terminal.
Category: Vibe coding | Difficulty: Beginner
Tools: Lovable (https://lovable.dev); v0 by Vercel (https://v0.app); Cloudflare Pages (https://pages.cloudflare.com)
Model: Built into the tool
MCP servers: none
Estimated monthly cost: $0–$25/mo. Lovable Free gives 5 build credits a day (30/month), enough for a simple page. Lovable Pro starts at $25/month if you need more. v0 Free gives 7 messages a day.
Config (brief.md):
# Landing page brief
Product: <one sentence>
Audience: <who>
One promise: <outcome they get>
Sections: hero, 3 benefits, how it works, FAQ, waitlist form
Style: dark, high contrast, one accent color
Must have: mobile-first, fast, no stock-photo people
Starter prompt: Build a single-page landing site from this brief: <paste brief.md>. Keep the copy short and concrete. Make the waitlist form store emails in a table, show a clear success message, and add basic spam protection. Then list anything I still need to set up before sharing the link.
Steps:
1. Write the brief first; it saves credits.
2. Paste it into Lovable (or v0 if you live in Next.js).
3. Iterate on copy, not layout, for the first few rounds.
4. Publish, then connect a custom domain if your plan allows it.
Watch out: Free-plan projects may show the builder's badge and have limits on private projects; check the plan page before launch.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: The $0 vibe-coding starter
URL: https://getagentloadout.com/loadouts/zero-dollar-starter/
Job: Learn to build and ship small apps without paying for anything.
Category: Vibe coding | Difficulty: Beginner
Tools: Google Antigravity (+ Antigravity CLI) (https://antigravity.google); Cloudflare Pages (https://pages.cloudflare.com); Supabase (https://supabase.com)
Model: Gemini 3.8 Flash (Antigravity free tier)
MCP servers: none
Estimated monthly cost: $0/mo. Antigravity's free tier has a weekly agent quota (Google doesn't publish a number). Cloudflare Pages and Supabase have free plans.
Config (AGENTS.md):
# Project rules
- Stack: plain HTML/CSS/JS or Vite. No paid services.
- Explain each change in one sentence before making it.
- Keep files small; one feature per commit.
- Never put secrets in client code.
Starter prompt: I'm new to coding. Build a small web app that <idea>. Use plain HTML, CSS and JavaScript (or Vite if needed). Explain what each file does in one line. When it works locally, give me the exact steps to deploy it to Cloudflare Pages for free.
Steps:
1. Install Antigravity CLI and sign in with a Google account.
2. Create a folder with the AGENTS.md above.
3. Run the starter prompt and build in small steps.
4. Deploy to Cloudflare Pages; add Supabase Free only if you need a database.
Watch out: Free quotas are capacity-based and can change. Supabase Free projects pause after a week of inactivity.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Local, private coding agent
URL: https://getagentloadout.com/loadouts/local-private-coding/
Job: Use an AI coding agent on code that must never leave your machine.
Category: Coding agents | Difficulty: Intermediate
Tools: Ollama (https://ollama.com); OpenCode (https://opencode.ai)
Model: Qwen3-Coder 30B via Ollama (qwen3-coder:30b)
MCP servers: none
Estimated monthly cost: $0/mo. Software is free. You need hardware: the qwen3-coder:30b download is 19 GB, so plan for a machine with enough RAM/VRAM to hold it.
Config (terminal):
ollama pull qwen3-coder:30b
# one-command launcher from the Ollama library page:
ollama launch opencode --model qwen3-coder
Starter prompt: You are running fully offline. Read the repository structure first and summarize it in 10 lines. Then <task>. Make small edits, run the tests after each change, and stop to ask if a change touches more than 3 files.
Steps:
1. Install Ollama and pull qwen3-coder:30b.
2. Install OpenCode, or use the ollama launch command.
3. Run the starter prompt inside your repo.
4. Keep tasks small; local models do best with tight scopes.
Watch out: Local models are slower and weaker than frontier cloud models on hard tasks. Ollama also lists a cloud variant (480b-cloud); don't use it if privacy is the point.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Personal research agent
URL: https://getagentloadout.com/loadouts/research-agent/
Job: Ask a question, get a sourced brief with links, in your terminal.
Category: Build agents | Difficulty: Intermediate
Tools: Claude Code (https://claude.com/product/claude-code); Exa (https://exa.ai); Firecrawl (https://www.firecrawl.dev)
Model: Claude Sonnet 5.5
MCP servers: Exa MCP (https://mcp.exa.ai/mcp, keyless with rate limits); Firecrawl MCP (npx -y firecrawl-mcp)
Estimated monthly cost: $20–$39/mo. Claude Pro $20. Exa's hosted MCP works without a key (rate limited). Firecrawl Free has 1,000 credits/month; Hobby is $19.
Config (.mcp.json):
{
  "mcpServers": {
    "exa": { "type": "http", "url": "https://mcp.exa.ai/mcp" },
    "firecrawl-mcp": {
      "command": "npx",
      "args": ["-y", "firecrawl-mcp"],
      "env": { "FIRECRAWL_API_KEY": "fc-YOUR_API_KEY" }
    }
  }
}
Starter prompt: Research: <question>. Search with Exa, read the 5-10 best primary sources with Firecrawl, and write research/<slug>.md with: a 5-bullet summary, key numbers with the source next to each, points where sources disagree, and a list of links. Mark anything you couldn't confirm as unverified.
Steps:
1. Add both MCP servers.
2. Run the prompt with one concrete question.
3. Save good outputs as templates for recurring research.
Watch out: Never paste API keys into chat. Keep them in the config or environment.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Deep research without code
URL: https://getagentloadout.com/loadouts/deep-research-no-code/
Job: Get a long, cited report on a topic with zero setup.
Category: Build agents | Difficulty: Beginner
Tools: Perplexity Pro (https://www.perplexity.ai/pro); ChatGPT (https://chatgpt.com)
Model: Built into each app
MCP servers: none
Estimated monthly cost: $17–$20/mo. Perplexity Pro is $17/month billed yearly; ChatGPT Plus is $20/month. Neither publishes exact Deep Research quotas.
Config (research-brief.md):
Topic:
What I already know:
What decision this informs:
Sources to prefer: primary docs, official pricing pages, papers
Sources to avoid: SEO listicles
Output: summary, table, open questions
Starter prompt: Run a deep research report on <topic>. I need it to decide <decision>. Prefer primary sources and official pages, include a comparison table, and end with what you could not verify.
Steps:
1. Fill in the brief.
2. Run it in one tool, then ask the other to fact-check the result.
3. Open the top 3 sources yourself before relying on numbers.
Watch out: Reports can cite weak sources; always check the key numbers.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Automatic PR code review
URL: https://getagentloadout.com/loadouts/pr-code-review/
Job: Get a thorough AI review on every pull request.
Category: Code review | Difficulty: Beginner
Tools: CodeRabbit (https://www.coderabbit.ai)
Model: Managed by CodeRabbit
MCP servers: none
Estimated monthly cost: $0–$30/mo. CodeRabbit Essentials is $30 per developer per month ($24 yearly); only developers who open PRs are billed.
Config (.coderabbit.yaml):
# unverified: check key names against CodeRabbit docs
reviews:
  profile: assertive
  path_instructions:
    - path: "src/**"
      instructions: "Flag missing auth checks, unvalidated input and secrets in code."
Starter prompt: @coderabbitai review this PR with a focus on security and error handling. Summarize the riskiest change first.
Steps:
1. Install the CodeRabbit app on your repo.
2. Add a config file with path-specific instructions.
3. Open a PR and reply to review comments to refine.
Watch out: AI reviewers miss business-logic bugs. Keep tests and a human read for risky changes.
Unverified: Config file keys not re-checked against CodeRabbit docs.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Review PRs from your terminal
URL: https://getagentloadout.com/loadouts/terminal-pr-review/
Job: Have your coding agent review, comment on and fix PRs on GitHub.
Category: Code review | Difficulty: Intermediate
Tools: Claude Code (https://claude.com/product/claude-code); GitHub Copilot (https://github.com/features/copilot)
Model: Claude Opus 5.5 for review depth
MCP servers: GitHub MCP (https://api.githubcopilot.com/mcp/)
Estimated monthly cost: $20/mo. Uses your Claude plan (Pro $20). Use a fine-grained GitHub token limited to the repos you need.
Config (terminal):
claude mcp add --transport http github https://api.githubcopilot.com/mcp/ \
  --header "Authorization: Bearer YOUR_GITHUB_PAT"
Starter prompt: Review PR #<number>. Check out the branch, run the tests, and read the diff. Post a review with: blocking issues, non-blocking suggestions, and questions. Don't push changes unless I say so.
Steps:
1. Create a fine-grained GitHub PAT.
2. Add the GitHub MCP server (command above).
3. Run /mcp to confirm it connected.
4. Run the prompt on a real PR.
Watch out: A PAT gives the agent whatever scopes you grant. Start read-only.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Browser automation agent (local)
URL: https://getagentloadout.com/loadouts/browser-automation-local/
Job: Let an agent click through websites: test your app, fill forms, collect data.
Category: Browser automation | Difficulty: Beginner
Tools: Claude Code (https://claude.com/product/claude-code); Playwright MCP (https://github.com/microsoft/playwright-mcp)
Model: Claude Sonnet 5.5
MCP servers: Playwright MCP (npx @playwright/mcp@latest)
Estimated monthly cost: $0–$20/mo. Playwright MCP is free and runs locally (Node.js 18+). Cost is your agent plan, e.g. Claude Pro $20.
Config (.mcp.json):
{
  "mcpServers": {
    "playwright": { "command": "npx", "args": ["@playwright/mcp@latest"] }
  }
}
Starter prompt: Open http://localhost:3000 and test the signup flow like a new user. Try one valid signup and three invalid ones. Report each step, what you expected, what happened, and take a screenshot of anything broken.
Steps:
1. Add the Playwright MCP server.
2. Start your app locally.
3. Run the prompt; add --headless to the args to hide the browser.
Watch out: Don't let it log into accounts that hold real money or data without watching.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Cloud browser agent at scale
URL: https://getagentloadout.com/loadouts/browser-agent-cloud/
Job: Run many browser sessions in the cloud for scraping or workflows.
Category: Browser automation | Difficulty: Advanced
Tools: Browserbase + Stagehand (https://www.browserbase.com)
Model: Your choice; Stagehand works with major model providers
MCP servers: none
Estimated monthly cost: $20–$99/mo. Browserbase Developer $20 (100 browser hours) or Startup $99 (500 hours). Model tokens are extra.
Config (agent.ts):
// unverified: check the current Stagehand API in its docs
import { Stagehand } from "@browserbasehq/stagehand";
const stagehand = new Stagehand({ env: "BROWSERBASE" });
await stagehand.init();
// await page.act("click the pricing link") ...
Starter prompt: Write a Stagehand script that opens <site>, finds <data>, and returns JSON matching this schema: <schema>. Retry once on failure and log each step.
Steps:
1. Create a Browserbase account and API key.
2. Install Stagehand.
3. Start with one page, then parallelize.
Watch out: Respect site terms and robots rules.
Unverified: Code snippet API not re-checked against current Stagehand docs.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Pay-per-minute browser agent
URL: https://getagentloadout.com/loadouts/browser-use-payg/
Job: Run an occasional browser agent without a monthly subscription.
Category: Browser automation | Difficulty: Intermediate
Tools: Browser Use (https://browser-use.com)
Model: Your choice (hosted agents add model cost + 20%)
MCP servers: none
Estimated monthly cost: $0–$5/mo. $0.02 per browser-hour billed by the minute; $5 minimum top-up; credits don't expire.
Config (task.txt):
Go to <site>. Find <thing>. Return a JSON list of {name, url, price}. Stop after 20 items.
Starter prompt: Using Browser Use, complete this task and return only the JSON: <task.txt>
Steps:
1. Top up $5.
2. Run a small task and check the cost.
3. Scale up once results are reliable.
Watch out: Residential proxy traffic costs $5/GB, which adds up on heavy pages.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Customer support agent (turnkey)
URL: https://getagentloadout.com/loadouts/support-agent-turnkey/
Job: Answer customer questions automatically in your existing helpdesk.
Category: Customer support | Difficulty: Beginner
Tools: Fin by Intercom (https://fin.ai)
Model: Managed by Intercom
MCP servers: none
Estimated monthly cost: From $49.50/mo. $0.99 per resolution with a 50-outcome monthly minimum (about $49.50/month at minimum). Scales with volume.
Config (guidance.md):
Tone: friendly, short answers.
Always: link the help article you used.
Hand off to a human when: refunds, legal, angry customer, anything about billing errors.
Never: promise features or dates.
Starter prompt: Use the guidance above. If you aren't sure the answer is in our help center, say so and offer a human.
Steps:
1. Connect Fin to your helpdesk.
2. Import help articles.
3. Write guidance and handoff rules.
4. Test on past tickets before going live.
Watch out: Busy months cost more; set alerts.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Website support chatbot
URL: https://getagentloadout.com/loadouts/support-bot-website/
Job: Put a chatbot trained on your docs on your site.
Category: Customer support | Difficulty: Beginner
Tools: Chatbase (https://www.chatbase.co)
Model: Managed by Chatbase
MCP servers: none
Estimated monthly cost: $40–$150/mo. Chatbase Hobby $40/month (700 message credits) or Standard $150 (4,000 credits).
Config (instructions.txt):
Answer only from the provided sources. If unsure, say "I'm not sure" and share the contact link. Keep answers under 80 words.
Starter prompt: Answer only from the provided sources. If the answer isn't there, give the contact link.
Steps:
1. Add your docs or URLs as sources.
2. Paste the instructions.
3. Embed the widget and review chat logs weekly.
Watch out: Branding removal is a $99/month add-on.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Build your own support agent
URL: https://getagentloadout.com/loadouts/support-agent-diy/
Job: Own the code and pick a cheap, fast model.
Category: Customer support | Difficulty: Advanced
Tools: Claude Agent SDK (https://docs.claude.com/en/api/agent-sdk/overview); OpenAI Agents SDK (https://openai.github.io/openai-agents-python/); Langfuse (https://langfuse.com)
Model: Claude Haiku 5.5 or GPT-6 Luna (both $0.10 in / $0.50 out per 1M tokens)
MCP servers: none
Estimated monthly cost: $0–$29/mo. API usage only, and small models are cheap (Haiku 5.5 / GPT-6 Luna: $0.10 per 1M input tokens). Langfuse Hobby is free; Core $29.
Config (system-prompt.md):
You are the support agent for <product>.
Tools: search_docs(query), get_order(id), escalate(reason).
Rules: answer from docs; call escalate for refunds or anything you can't verify; never invent order data.
Starter prompt: Build a support agent with the tools in system-prompt.md. Add tracing with Langfuse, an eval set of 30 real past questions, and a script that reports accuracy before each deploy.
Steps:
1. Pick one SDK.
2. Implement three tools.
3. Add tracing and an eval set.
4. Ship behind a 'talk to a human' button.
Watch out: You own uptime and safety. Keep escalation easy.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Discord bot in an afternoon
URL: https://getagentloadout.com/loadouts/discord-bot/
Job: Build and host an AI-powered Discord bot for your server.
Category: Build agents | Difficulty: Intermediate
Tools: Claude Code (https://claude.com/product/claude-code); discord.js (https://discord.js.org); Railway (https://railway.com)
Model: Claude Sonnet 5.5 to build; Claude Haiku 5.5 inside the bot
MCP servers: none
Estimated monthly cost: $25–$30/mo. Claude Pro $20 to build, Railway Hobby $5 (includes $5 usage). Bot API calls are pay-per-token on top, small with Haiku 5.5.
Config (.env.example):
DISCORD_TOKEN=
ANTHROPIC_API_KEY=
# never commit .env
Starter prompt: Create a discord.js bot with slash commands /ask (answers with Claude Haiku 5.5, max 300 tokens) and /help. Add per-user rate limiting, ignore other bots, read secrets from .env, and add a Procfile or start script so it runs on Railway.
Steps:
1. Create the app in the Discord Developer Portal and copy the token.
2. Run the prompt in Claude Code.
3. Test in a private server.
4. Deploy to Railway with the env vars set.
Watch out: Rate limit /ask or a busy server will run up your API bill.
Unverified: discord.js details not re-checked this cycle.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Analyze a spreadsheet in plain English
URL: https://getagentloadout.com/loadouts/data-analysis-no-code/
Job: Upload a CSV or connect a database and get charts and answers.
Category: Data analysis | Difficulty: Beginner
Tools: Julius AI (https://julius.ai); ChatGPT (https://chatgpt.com)
Model: Built into each app
MCP servers: none
Estimated monthly cost: $0–$20/mo. Julius Free, or Plus $20/month (2,000 credits). ChatGPT Plus $20 also does file analysis.
Config (question.md):
Data: sales_2026.csv
Question: which channels grew fastest month over month?
Output: one chart, a 5-line summary, and the code you used
Starter prompt: Here's my data. First describe the columns and any data quality issues. Then answer: <question>. Show one chart and the code you used.
Steps:
1. Upload the file.
2. Ask for a data quality check first.
3. Ask your question and keep the generated code.
Watch out: Don't upload sensitive data without checking the vendor's data policy.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Agent-driven data notebook
URL: https://getagentloadout.com/loadouts/data-analysis-notebook/
Job: Do reproducible analysis in Python with an agent writing the code.
Category: Data analysis | Difficulty: Intermediate
Tools: Claude Code (https://claude.com/product/claude-code); OpenAI Codex (https://openai.com/codex)
Model: Claude Sonnet 5.5 or GPT-6.1 Sol
MCP servers: none
Estimated monthly cost: $20/mo. Claude Pro or ChatGPT Plus at $20/month.
Config (AGENTS.md):
- Use Python with pandas and matplotlib in a virtualenv.
- Put analysis in notebooks/ and reusable code in src/.
- Never modify files in data/raw/.
- Every chart gets a one-line takeaway.
Starter prompt: Load data/raw/<file>. Check for missing values and outliers, then answer <question>. Write the work as a notebook I can re-run, and finish with 3 takeaways.
Steps:
1. Create a repo with data/raw and AGENTS.md.
2. Run the prompt.
3. Commit the notebook so results are reproducible.
Watch out: Check outputs against a manual spot-check before sharing.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: AI workflow automation (no code)
URL: https://getagentloadout.com/loadouts/workflow-automation/
Job: Automate a business process: triage email, enrich leads, post updates.
Category: Automation | Difficulty: Beginner
Tools: n8n (https://n8n.io)
Model: Any model via n8n's AI nodes (bring your own API key)
MCP servers: none
Estimated monthly cost: Usage-based. n8n Cloud Starter €24/month (€20 billed yearly), 2,500 executions. Self-hosting is also possible. Model API usage is extra.
Config (workflow.md):
Trigger: new form submission
Step 1: AI agent classifies lead (hot/warm/cold) with a reason
Step 2: enrich with company website summary
Step 3: post to Slack and add to sheet
Starter prompt: Build this n8n workflow: <workflow.md>. Use an AI Agent node with a short system prompt, return JSON, and add an error branch that notifies me.
Steps:
1. Sketch the workflow in plain text.
2. Build it in n8n (its Assistant can help).
3. Run 10 test items before turning it on.
Watch out: Assistant credits were revalued on Oct 1, 2026; plans now show fewer, higher-value credits.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Delegate tasks to a background agent
URL: https://getagentloadout.com/loadouts/async-cloud-agent/
Job: Hand off small GitHub tasks and review the PR later.
Category: Coding agents | Difficulty: Beginner
Tools: Jules (https://jules.google); OpenAI Codex (https://openai.com/codex)
Model: Gemini (Jules) or GPT-6.1 Sol (Codex)
MCP servers: none
Estimated monthly cost: $0–$20/mo. Jules Free gives 15 tasks per 24h. Codex comes with ChatGPT plans (Plus $20).
Config (task-template.md):
Goal:
Files likely involved:
Done when: tests pass and <check>
Out of scope:
Starter prompt: Task: <goal>. Likely files: <files>. Done when tests pass and <check>. Open a PR with a summary of what you changed and why.
Steps:
1. Connect your GitHub repo.
2. Write tasks with a clear 'done when'.
3. Review PRs like a teammate's.
Watch out: Small, well-specified tasks only. Vague tasks produce big, hard-to-review PRs.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Assign GitHub issues to an agent
URL: https://getagentloadout.com/loadouts/issue-to-pr-copilot/
Job: Turn issues into draft PRs without leaving GitHub.
Category: Coding agents | Difficulty: Beginner
Tools: GitHub Copilot (https://github.com/features/copilot)
Model: Auto model selection or premium models (plan-dependent)
MCP servers: none
Estimated monthly cost: $10–$39/mo. Copilot Pro $10 (1,500 AI credits/month) or Pro+ $39 (7,000). Agent usage consumes AI credits.
Config (.github/copilot-instructions.md):
- Run `npm test` before opening a PR.
- Follow the existing folder structure.
- Keep PRs under 300 changed lines; split otherwise.
Starter prompt: (In the issue) Acceptance criteria: <list>. Tests to add: <list>. Assign to Copilot.
Steps:
1. Add repository instructions.
2. Write issues with acceptance criteria.
3. Assign to Copilot and review the draft PR.
Watch out: Since June 2026 usage is credit-based; heavy agent use can exhaust a Pro plan.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Cheapest paid IDE agent
URL: https://getagentloadout.com/loadouts/budget-ide-agent/
Job: Get unlimited completions plus an agent for the price of a lunch.
Category: Coding agents | Difficulty: Beginner
Tools: GitHub Copilot (https://github.com/features/copilot); Cursor (https://cursor.com)
Model: Copilot auto/selection or Cursor's own models
MCP servers: none
Estimated monthly cost: $8–$10/mo. Cursor Plus $8 (Cursor models only, no cloud agents) or Copilot Pro $10 (unlimited completions + 1,500 AI credits).
Config (AGENTS.md):
- Prefer small diffs.
- Ask before adding dependencies.
- Run the linter after edits.
Starter prompt: Explain this codebase's structure in 10 lines, then implement <small feature> with a test.
Steps:
1. Pick one editor.
2. Add AGENTS.md.
3. Use the agent for small features; completions for the rest.
Watch out: Cursor Plus doesn't include third-party models (Claude, GPT); you need Premium ($30) and prepaid credit for those.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Cursor with Claude or GPT models
URL: https://getagentloadout.com/loadouts/cursor-frontier-models/
Job: Use Cursor's editor with frontier third-party models after the Oct 2026 plan change.
Category: Coding agents | Difficulty: Intermediate
Tools: Cursor (https://cursor.com)
Model: Claude Opus 5.5 / GPT-6.1 Sol at API list price
MCP servers: none
Estimated monthly cost: From $30/mo. Cursor Premium $30/month plus prepaid credit for third-party models at API list price (e.g. Claude Sonnet 5.5 $2/$10 per 1M tokens).
Config (.cursor/rules/project.mdc):
---
description: project rules
alwaysApply: true
---
- Use TypeScript strict mode.
- Ask before editing more than 5 files.
Starter prompt: Plan first: list the files you'll touch and why. Then implement <feature> and run the tests.
Steps:
1. Choose Premium or above.
2. Add prepaid credit.
3. Pick the model per task; use Cursor models for cheap edits.
Watch out: Legacy Pro/Pro+ users keep their old plans; switching to The X Subscription is permanent.
Unverified: Cursor rules file format not re-checked this cycle.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Phone voice agent
URL: https://getagentloadout.com/loadouts/voice-agent/
Job: Answer calls or book appointments with a voice AI.
Category: Build agents | Difficulty: Intermediate
Tools: Vapi (https://vapi.ai); ElevenLabs Agents (https://elevenlabs.io/agents)
Model: Fast, cheap LLM (e.g. GPT-6 Luna or Claude Haiku 5.5)
MCP servers: none
Estimated monthly cost: Usage-based. Usage-based: Vapi $0.05/min platform fee plus model, speech and telephony at cost; ElevenLabs Agents $0.08/min plus LLM.
Config (assistant.md):
Role: receptionist for <business>.
Goals: answer hours/location, book appointments, take messages.
Rules: confirm name and phone number back; transfer to a human for emergencies.
Starter prompt: You are the receptionist described in assistant.md. Keep replies under two sentences. Confirm details before booking.
Steps:
1. Write the assistant spec.
2. Build in one platform and attach a phone number.
3. Test 20 calls, including rude and confusing callers.
Watch out: Telephony and model costs are on top of the platform fee.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Vibe-code a mobile app
URL: https://getagentloadout.com/loadouts/mobile-app/
Job: Build an iOS/Android app with React Native and ship test builds.
Category: Vibe coding | Difficulty: Intermediate
Tools: Claude Code (https://claude.com/product/claude-code); Expo (EAS) (https://expo.dev); Supabase (https://supabase.com)
Model: Claude Sonnet 5.5
MCP servers: Supabase MCP (optional, for the backend)
Estimated monthly cost: $20–$45/mo. Claude Pro $20. EAS Free includes 15 Android + 15 iOS builds/month; Starter $19. Supabase Pro $25 once you go live. App store developer fees are extra.
Config (terminal):
npx create-expo-app@latest my-app
cd my-app
claude
Starter prompt: This is an Expo app. Build <app idea> with Expo Router: 3 screens, local state first, then Supabase for sync. After each screen, tell me how to test it on my phone.
Steps:
1. Scaffold with create-expo-app.
2. Build screen by screen with the agent.
3. Use EAS to make test builds.
Watch out: App store review and developer accounts take time; start early.
Unverified: create-expo-app command not re-checked this cycle.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Vibe-code a browser game
URL: https://getagentloadout.com/loadouts/browser-game/
Job: Prototype a small 2D game and share it with a link.
Category: Vibe coding | Difficulty: Beginner
Tools: Claude Code (https://claude.com/product/claude-code); Phaser (https://phaser.io); Cloudflare Pages (https://pages.cloudflare.com)
Model: Claude Sonnet 5.5
MCP servers: Playwright MCP (to playtest)
Estimated monthly cost: $0–$20/mo. Claude Pro $20 (or the $0 starter stack). Cloudflare Pages hosting is free.
Config (GAME.md):
Genre: top-down arena
Core loop: dodge, collect, upgrade
Controls: WASD + mouse
One-sentence fantasy: <...>
Scope: 1 level, 3 enemy types, a score screen
Starter prompt: Build the game in GAME.md with Phaser and Vite. Start with movement and one enemy, then stop so I can playtest. Keep tuning values (speed, health, spawn rate) in one config file.
Steps:
1. Write GAME.md.
2. Build the core loop first.
3. Playtest, tweak the config, repeat.
4. Deploy to Cloudflare Pages and share the link.
Watch out: Use assets you have rights to.
Unverified: Phaser details not re-checked this cycle.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: See what your agent is doing
URL: https://getagentloadout.com/loadouts/agent-observability/
Job: Trace, debug and evaluate an agent in production.
Category: Build agents | Difficulty: Intermediate
Tools: Langfuse (https://langfuse.com)
Model: Any
MCP servers: none
Estimated monthly cost: $0–$29/mo. Langfuse Hobby free (50k units/month), Core $29 (100k units). Open source and self-hostable.
Config (env):
LANGFUSE_PUBLIC_KEY=
LANGFUSE_SECRET_KEY=
LANGFUSE_HOST=
Starter prompt: Add Langfuse tracing to every model call and tool call in this agent. Tag traces with user id and version. Then write an eval script that replays 20 saved traces and reports regressions.
Steps:
1. Add keys.
2. Instrument model and tool calls.
3. Review failing traces weekly.
Watch out: Don't log personal data you don't need.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Build your own MCP server
URL: https://getagentloadout.com/loadouts/build-mcp-server/
Job: Give AI assistants access to your app or data through MCP.
Category: Build agents | Difficulty: Advanced
Tools: Claude Code (https://claude.com/product/claude-code); MCP Inspector (https://github.com/modelcontextprotocol/inspector)
Model: Claude Sonnet 5.5
MCP servers: none
Estimated monthly cost: $0–$20/mo. Open-source SDK and Inspector are free; cost is your agent plan.
Config (terminal):
# test any MCP server visually
npx @modelcontextprotocol/inspector
Starter prompt: Create a TypeScript MCP server with the official SDK that exposes these tools: <tools>. Validate inputs, return clear errors, add a README with a config snippet for Claude Code and Cursor, and test it with the MCP Inspector.
Steps:
1. Define 2-3 tools with clear names.
2. Build with the agent.
3. Test in the Inspector, then in a real client.
Watch out: Treat every tool as an API: authenticate it and limit what it can do.
Unverified: Inspector command and SDK details not re-checked this cycle.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Turn websites into clean data
URL: https://getagentloadout.com/loadouts/scrape-to-data/
Job: Extract structured data from many pages without writing a scraper.
Category: Build agents | Difficulty: Intermediate
Tools: Firecrawl (https://www.firecrawl.dev); Claude Code (https://claude.com/product/claude-code)
Model: Claude Sonnet 5.5
MCP servers: Firecrawl MCP
Estimated monthly cost: $0–$39/mo. Firecrawl Free (1,000 credits/month) or Hobby $19. Plus your agent plan.
Config (.mcp.json):
{
  "mcpServers": {
    "firecrawl-mcp": {
      "command": "npx",
      "args": ["-y", "firecrawl-mcp"],
      "env": { "FIRECRAWL_API_KEY": "fc-YOUR_API_KEY" }
    }
  }
}
Starter prompt: Map <site>, then extract every <item> page into data/items.json with fields {name, url, price, updated}. Deduplicate, validate the JSON, and report pages that failed.
Steps:
1. Add the Firecrawl MCP.
2. Try one page first.
3. Scale to the full site.
Watch out: Check site terms before scraping.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Pre-launch security check for a vibe-coded app
URL: https://getagentloadout.com/loadouts/prelaunch-security-check/
Job: Catch leaked keys, missing auth and open database tables before you launch.
Category: Vibe coding | Difficulty: Intermediate
Tools: Claude Code (https://claude.com/product/claude-code); Supabase (https://supabase.com)
Model: Claude Opus 5.5
MCP servers: Supabase MCP in read-only mode
Estimated monthly cost: $20/mo. Claude Pro $20. Supabase MCP read-only access is free.
Config (.mcp.json):
{
  "mcpServers": {
    "supabase": { "type": "http", "url": "https://mcp.supabase.com/mcp?project_ref=YOUR_PROJECT_REF&read_only=true" }
  }
}
Starter prompt: Do a pre-launch security review. Check: secrets in client code or git history, API routes without auth, Supabase tables without RLS or with overly broad policies, missing rate limits, and unvalidated input. Output a table: issue, file, severity, fix. Don't change anything yet.
Steps:
1. Connect Supabase MCP with read_only=true.
2. Run the prompt.
3. Fix critical issues one at a time, then re-run.
Watch out: This doesn't replace a professional audit for apps holding sensitive data.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Production agent in TypeScript
URL: https://getagentloadout.com/loadouts/ts-production-agent/
Job: Build an agent with tools, memory and workflows you can deploy.
Category: Build agents | Difficulty: Advanced
Tools: Mastra (https://mastra.ai); Langfuse (https://langfuse.com); Claude Code (https://claude.com/product/claude-code)
Model: GPT-6.1 Sol or Claude Sonnet 5.5 ($2 in / $10 out per 1M tokens)
MCP servers: none
Estimated monthly cost: $0–$29/mo. Mastra is open source. Model usage is pay-per-token. Langfuse Hobby free, Core $29.
Config (agent-plan.md):
Agent: <name>
Tools: <3-5 tools>
Memory: per-user thread
Workflow: plan -> act -> verify -> report
Evals: 25 cases in evals/
Starter prompt: Using Mastra, implement the agent in agent-plan.md with typed tools, a workflow with a human-approval step before any write action, and Langfuse tracing. Add the eval cases and a CI script.
Steps:
1. Write the plan.
2. Build tools first, then the workflow.
3. Add evals before adding features.
Watch out: Pick LangGraph instead if your team is Python-first.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: AI agent for P&L and month-end reporting
URL: https://getagentloadout.com/loadouts/finance-pl-reporting/
Job: Pull your real profit and loss from Xero or QuickBooks, explain the variances, and build the management report in Excel.
Category: Finance & reporting | Difficulty: Intermediate
Tools: Claude (app) (https://claude.com); Julius AI (https://julius.ai)
Model: Claude Sonnet 5.5 (default) / Opus 5.5 for variance analysis
MCP servers: Xero MCP server, official (https://github.com/XeroAPI/xero-mcp-server); QuickBooks Online MCP server, official (https://github.com/intuit/quickbooks-online-mcp-server)
Estimated monthly cost: $20–$40/mo. Claude Pro $20/month ($17 billed yearly) covers the Claude app, MCP servers in Claude Desktop and Claude for Excel. Add Julius Plus $20/month only if you work from CSV exports. Your Xero or QuickBooks subscription is not included.
Config (claude_desktop_config.json):
{
  "mcpServers": {
    "xero": {
      "command": "npx",
      "args": ["-y", "@xeroapi/xero-mcp-server@latest"],
      "env": {
        "XERO_CLIENT_ID": "your_client_id_here",
        "XERO_CLIENT_SECRET": "your_client_secret_here",
        "XERO_SCOPES": "accounting.reports.profitandloss.read accounting.reports.balancesheet.read accounting.reports.trialbalance.read accounting.invoices.read accounting.settings.read"
      }
    }
  }
}
Starter prompt: Using the Xero tools, pull the profit and loss for last month and for the same month last year. Build a one-page management report: revenue, gross margin, operating expenses by category and net profit, with the change versus last year in currency and %. For every line that moved more than 10%, give a one-line likely reason and the invoices or accounts behind it. Read only: do not create or edit anything in Xero. List any number you could not tie back to the P&L report.
Steps:
1. Create a Xero Custom Connection for your organisation (or an Intuit developer app for QuickBooks). Try it on the Xero Demo Company first.
2. In Claude Desktop, open Settings > Developer > Edit config, paste the config with your client ID and secret, and restart.
3. Run the starter prompt, then tie the totals back to the P&L report in Xero before you trust anything else.
4. Install Claude for Excel from Microsoft AppSource and ask Claude to turn the report into a workbook with live formulas.
5. No accounting connection? Export the P&L or general ledger to CSV and ask the same questions in Julius.
Watch out: Both servers can also write (the Xero one creates invoices and contacts), so grant read-only scopes. For QuickBooks, set QUICKBOOKS_DISABLE_WRITE, QUICKBOOKS_DISABLE_UPDATE and QUICKBOOKS_DISABLE_DELETE to true. AI can flip signs or mix up periods: check totals against the source report. This helps you prepare reports; it doesn't replace your accountant.
Unverified: The read-only Xero scope list in the config hasn't been tested on a live custom connection. Xero moved to granular scopes for connections created from Apr 29, 2026; if a tool fails, add the scope it asks for. Xero custom connection and QuickBooks developer app costs, if any, aren't included in the estimate.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Personal AI agent for everyday life
URL: https://getagentloadout.com/loadouts/personal-ai-assistant/
Job: Get a daily brief of your calendar, the emails that actually need you, and what's due this week, with replies drafted but never sent without you.
Category: Personal assistants | Difficulty: Beginner
Tools: ChatGPT (https://chatgpt.com); Claude (app) (https://claude.com); Lindy (https://www.lindy.ai)
Model: Built into each app
MCP servers: none
Estimated monthly cost: $20–$70/mo. ChatGPT Plus $20/month runs the scheduled brief (5 active tasks, hourly schedules). Lindy Team is $49.99/month per user for 5,000 credits if you want inbox triage and meeting scheduling all day; try its free $50 credits (7 days) first. Claude Pro ($20) is an alternative if you only need answers about your email.
Config (assistant-brief.md):
# About me
Time zone:
Work hours:
People who always get a reply:
Never do without asking: send email, accept invites, pay, delete

# Daily brief (weekdays 7:30)
- Today's meetings, with a one-line prep note each
- Emails from real people that need a reply, with a short draft
- Bills, deliveries and renewals due this week

# Weekly (Sunday 18:00)
- Next week at a glance, with conflicts flagged
Starter prompt: Every weekday at 7:30, check my Gmail and Google Calendar and send me a brief: today's meetings with a one-line prep note each, emails from real people that need a reply (skip newsletters and receipts), and anything due this week such as bills, deliveries or renewals. Draft replies, but never send email, accept invites or delete anything without asking me first. Here's my brief: <paste assistant-brief.md>
Steps:
1. Fill in the brief, especially the 'never do without asking' line.
2. In ChatGPT, connect Gmail and Google Calendar under Settings > Apps.
3. Paste the starter prompt to create a scheduled task, then check it on the Scheduled page.
4. Want it to react when an email arrives? Create an event-triggered task in Work (Plus and above, Gmail, Slack or GitHub).
5. After two weeks, if you're still triaging email by hand, trial Lindy for inbox management and meeting scheduling.
Watch out: ChatGPT pauses a task when an action that sends a message or changes data needs approval; keep it that way. Claude's Gmail and Calendar connectors are read-only, which is a safe first step. Never put passwords, account numbers or health details in task instructions, since shared task links include the full instructions.
Unverified: Which Google Calendar actions ChatGPT can take (not just read) depends on your account and the app's permissions; this loadout only relies on reading.
Last checked: 2026-10-21. Hands-on tested: no

## Loadout: Build a full app with login and payments, no code
URL: https://getagentloadout.com/loadouts/no-code-full-stack-app/
Job: Go from an idea to a published web app with sign-up, a database and a paid plan, without opening a terminal.
Category: Vibe coding | Difficulty: Beginner
Tools: Lovable (https://lovable.dev); Stripe (https://stripe.com)
Model: Built into Lovable
MCP servers: none
Estimated monthly cost: $0–$25/mo. Lovable Free (5 build credits a day, 30 a month) is enough to try it and can take payments through your own Stripe account. Built-in payments need Pro, from $25/month ($21 billed yearly). Hosting and the backend use run credits; Free, Pro and Business currently include a 20-credit monthly Cloud grant (a temporary offer). Stripe charges 2.9% + 30¢ per successful US card payment.
Config (app-spec.md):
# App spec
Who it's for:
The one job it does:
Pages: landing, sign up / log in, dashboard, settings, pricing
Data: <each table, and what one row is>
Who can see what: users only see their own rows
Paid plan: <name>, <price>/month, unlocks <feature>
Out of scope for v1:
Starter prompt: Build this app from the spec below on the built-in backend (Cloud). Start with email sign-in and the dashboard. Every table needs row-level security so users only see their own data: show me the tables and access rules before you build pages. Then add the paid plan with built-in payments in test mode. After each step, tell me exactly what to click in the preview to test it. <paste app-spec.md>
Steps:
1. Write the spec and keep v1 to one job; it saves credits.
2. Paste the starter prompt into Lovable and approve the tables and access rules before any pages get built.
3. Test sign-up and checkout in the preview with test cards; no real money moves there.
4. Run the Deep security scan in the Security view and fix every Critical finding.
5. Publish (a Quick scan runs automatically), then finish your payment provider's verification to go live.
Watch out: The built-in scans catch common mistakes but don't replace a real review if you handle sensitive data. A project uses either built-in payments or your own Stripe account, not both. Selling physical goods? Lovable points you to its Shopify integration instead.
Last checked: 2026-10-21. Hands-on tested: no

## Tier list: AI coding agents
S: Claude Code — Strong agent loop, first-class MCP support, included in a $20 plan. | OpenAI Codex — Bundled with ChatGPT plans (Plus $20), runs locally and in the cloud, GPT-6.1 Sol.
A: Cursor — Best-known AI editor; Plus at $8 is the cheapest entry, but third-party models now cost extra (Premium+). | GitHub Copilot — $10 Pro with unlimited completions and an agent where your issues live; usage now credit-based. | Devin Desktop (formerly Windsurf) — Former Windsurf; manages local and cloud agents from one place.
B: OpenCode — Free, open source, any provider or local model. | Cline — Free, open source, popular in VS Code. | Google Antigravity (+ Antigravity CLI) — Free tier with a weekly quota; replaced Gemini CLI for consumers. | Jules — Free async tasks (15/day); good for small chores.
C: Aider — Solid and git-native, but less agentic than the tools above.
Method: Editorial ranking from official docs, pricing and public release notes. Not yet hands-on tested.

## Tier list: Vibe-coding app builders
S: Lovable — Hosting, database and auth built in; free daily credits.
A: v0 by Vercel — Best if you deploy on Vercel and use Next.js; 7 free messages/day. | Replit — Full IDE plus agent; Core $20 includes $25 credits.
B: Bolt.new — Capable in-browser builder; pricing not confirmed this cycle.
C: —
Method: Editorial ranking from official docs, pricing and public release notes. Not yet hands-on tested.

## Tier list: AI code review
S: CodeRabbit — Purpose-built reviewer; only PR authors are billed.
A: GitHub Copilot — Native to GitHub; consumes AI credits and Actions minutes. | Cursor — Bugbot included on Premium ($30) and above.
B: Claude Code — Deep reviews on demand via the GitHub MCP, but you have to run them yourself.
C: —
Method: Editorial ranking from official docs, pricing and public release notes. Not yet hands-on tested.

## Tier list: Browser automation for agents
S: Playwright MCP — Free, local, works in any MCP client.
A: Browserbase + Stagehand — Hosted browsers at scale with an open-source SDK (Stagehand). | Browser Use — Pay-per-minute cloud browsers, no subscription.
B: —
C: —
Method: Editorial ranking from official docs, pricing and public release notes. Not yet hands-on tested.

## Tier list: Agent frameworks
S: LangGraph — Durable, checkpointed workflows with human approval (Python/TS). | Mastra — All-in-one TypeScript stack: agents, workflows, memory, evals.
A: Claude Agent SDK — The Claude Code harness as an SDK; Claude-only. | OpenAI Agents SDK — Lightweight handoffs and guardrails; OpenAI-first.
B: n8n — No-code agents inside visual workflows.
C: —
Method: Editorial ranking from official docs, pricing and public release notes. Not yet hands-on tested.

## Tier list: Web search and scraping for agents
S: Exa — Search built for AI; hosted MCP works without a key. | Firecrawl — Any URL to clean markdown; 1,000 free credits/month.
A: Tavily — Simple credit pricing; 1,000 free credits/month.
B: —
C: —
Method: Editorial ranking from official docs, pricing and public release notes. Not yet hands-on tested.

## Tier list: Customer support agents
S: Fin by Intercom — Pay per resolution ($0.99), works with existing helpdesks.
A: Chatbase — Fastest way to put a docs-trained bot on a website.
B: Claude Agent SDK — Build your own for full control; more work.
C: —
Method: Editorial ranking from official docs, pricing and public release notes. Not yet hands-on tested.

## Patch Notes
- 2026-10-15 [NEW] Agentic autofix for code scanning alerts is GA: Assign a CodeQL alert to Copilot; it explores the codebase, proposes a fix, reruns CodeQL to confirm and opens a draft PR. Enterprise Cloud with Code Security / Advanced Security only; consumes AI Credits and Actions minutes. Source: https://github.blog/changelog/2026-10-15-agentic-autofix-for-code-scanning-alerts-general-availability/
- 2026-10-13 [BUFF] Gemini CLI v0.64.0: stability release: Atomic file writes with state recovery, @file:line references, and Ctrl+C / terminal UI fixes. Reminder: free and Google AI Pro/Ultra users were moved to Antigravity CLI on June 18, 2026; Gemini CLI remains for paid API keys and enterprise. Source: https://geminicli.com/docs/changelogs/latest/
- 2026-10-12 [CHANGE] Cursor joins The X Subscription: New plans: Plus $8, Premium $30, Super $100, Ultra $200. Buff: a cheaper $8 entry point and larger usage for Cursor models. Nerf: third-party models (Claude, GPT) are no longer included; on Premium and above they're billed at API list price from prepaid credit. Legacy Pro/Pro+/Ultra continue for existing subscribers. Source: https://cursor.com/help/account-and-billing/pricing
- 2026-10-07 [NEW] Claude Haiku 5.5 released: Anthropic's small model: $0.10 input / $0.50 output per 1M tokens for prompts up to 100K tokens ($0.50 / $2.50 above). A strong default for high-volume agents like support bots. Source: https://www.anthropic.com/claude-haiku-5-5
- 2026-10-01 [CHANGE] n8n Assistant credits revalued: Each Assistant credit is now worth 1 cent, so plans show fewer credits (Starter 2,300 to 1,600; Pro-1 5,700 to 4,000; Pro-2 13,700 to 9,600). n8n says each request uses fewer credits, so capacity is about the same. Workflow runs don't use Assistant credits. Source: https://support.n8n.io/article/fewer-assistant-credits-october-2026
- 2026-09-29 [NEW] OpenAI releases GPT-6.1 Sol: Available in the API (gpt-6.1-sol) and in Codex. $2 input / $10 output per 1M tokens, cached input $0.10. Source: https://openai.com/index/introducing-gpt-6-1-sol/
- 2026-09-28 [NEW] Claude Sonnet 5.5 released: $2 input / $10 output per 1M tokens. The default model in most of our Claude Code loadouts. Source: https://www.anthropic.com/claude-sonnet-5-5
- 2026-09-22 [NEW] Claude Opus 5.5 released: Anthropic's top model at $4 input / $20 output per 1M tokens. Use it for planning and reviews. Source: https://www.anthropic.com/claude-opus-5-5

## Tool pricing
- Claude Code (https://claude.com/product/claude-code): Pro $20 ($17/mo billed yearly; 5-hour and weekly usage limits); Max 5x $100 (5x Pro usage per session); Max 20x $200 (20x Pro usage per session). Status: verified. Source: https://claude.com/pricing?mode=code
- OpenAI Codex (https://openai.com/codex): Free $0 (Lightweight tasks, GPT-6 Luna); Go $8 (Lightweight tasks); Plus $20 (GPT-6.1 Sol and GPT-6 Luna); Pro $100 (Also $200 and $500 tiers). Status: verified. Source: https://modelpick.dev/coding-plans/
- Cursor (https://cursor.com): Free $0 (Limited usage); Plus $8 (Cursor models only (Grok 4.7, Composer 2.5); no cloud agents or third-party models); Premium $30 (4x Plus usage; cloud agents, Bugbot, third-party models at API list price); Super $100 (15x Plus usage); Ultra $200 (50x Plus usage). Status: verified. Source: https://cursor.com/help/account-and-billing/pricing
- GitHub Copilot (https://github.com/features/copilot): Free $0 (2,000 completions/month, limited chat and agent); Pro $10 (1,500 AI credits/month (1,000 base + 500 flex)); Pro+ $39 (7,000 AI credits/month); Max $100 (20,000 AI credits/month). Status: verified. Source: https://docs.github.com/en/copilot/get-started/plans
- Devin Desktop (formerly Windsurf) (https://devin.ai): Free $0; Pro $20 (Quota-based daily/weekly allowances); Max $200. Status: verified. Source: https://devin.ai/pricing
- Google Antigravity (+ Antigravity CLI) (https://antigravity.google): Free $0 (Weekly agent quota; Gemini 3.8 Flash as core model); Google AI Pro $19.99 (Higher quota refreshed every 5 hours). Status: verified. Source: https://antigravity.google/docs/plans/
- Jules (https://jules.google): Free $0 (15 tasks per rolling 24h, 3 concurrent); Google AI Pro $19.99 (100 tasks/day, 15 concurrent); Google AI Ultra $99.99 (From $99.99; 300 tasks/day). Status: verified. Source: https://jules.google/docs/usage-limits/
- OpenCode (https://opencode.ai): Open source $0 (Pay only for model usage). Status: verified. Source: https://opencode.ai/docs/providers/
- Cline (https://cline.bot): Open source $0 (Pay only for model usage). Status: verified. Source: https://github.com/cline/cline
- Aider (https://aider.chat): Open source $0 (Pay only for model usage). Status: verified. Source: https://aider.chat/docs/llms.html
- Ollama (https://ollama.com): Local $0 (Free to run on your own hardware). Status: verified. Source: https://ollama.com/library/qwen3-coder
- Lovable (https://lovable.dev): Free $0 (5 daily build credits, capped at 30/month); Pro $25 (From $25/mo ($21/mo yearly), from 100 credits/month); Business $50 (From $50/mo, SSO and governance). Status: verified. Source: https://docs.lovable.dev/introduction/subscription-plans
- v0 by Vercel (https://v0.app): Free $0 (7 messages/day); Paid $30 (Per user; $30 monthly credits plus $2 daily credits); Business $100 (Per user). Status: verified. Source: https://v0.app/pricing
- Replit (https://replit.com): Core $20 ($18/mo yearly; includes $25 monthly credits); Pro $100 ($95/mo yearly). Status: verified. Source: https://replit.com/pricing
- Bolt.new (https://bolt.new): Pro (Price not confirmed). Status: unverified. Source: https://bolt.new/pricing
- Supabase (https://supabase.com): Free $0 (500 MB DB, 50k MAU, pauses after 1 week inactive); Pro $25 (Includes one Micro compute project; usage beyond quotas billed). Status: verified. Source: https://supabase.com/pricing
- Vercel (https://vercel.com): Hobby $0 (Personal, non-commercial use); Pro $20 (One deploying seat plus $20 usage credit). Status: verified. Source: https://vercel.com/pricing
- Cloudflare Pages (https://pages.cloudflare.com): Free $0 (Unlimited sites and bandwidth, 500 builds/month). Status: verified. Source: https://developers.cloudflare.com/pages/platform/limits/
- Railway (https://railway.com): Hobby $5 (Includes $5 of usage; overage billed on top). Status: verified. Source: https://docs.railway.com/pricing/plans
- Stripe (https://stripe.com): Pay per transaction (No monthly fee; 2.9% + 30¢ per successful US card charge). Status: verified. Source: https://stripe.com/pricing
- Expo (EAS) (https://expo.dev): Free $0 (15 Android + 15 iOS builds/month); Starter $19 (Includes $45 build credit). Status: verified. Source: https://expo.dev/pricing
- CodeRabbit (https://www.coderabbit.ai): Essentials $30 (Per developer; $24/dev/mo yearly. Only PR authors are billed). Status: verified. Source: https://www.coderabbit.ai/pricing
- Playwright MCP (https://github.com/microsoft/playwright-mcp): Open source $0 (Runs locally with npx). Status: verified. Source: https://playwright.dev/docs/getting-started-mcp
- Browserbase + Stagehand (https://www.browserbase.com): Free $0; Developer $20 (100 browser hours, 25 concurrent); Startup $99 (500 browser hours, 100 concurrent). Status: verified. Source: https://docs.browserbase.com/account/billing/plans
- Browser Use (https://browser-use.com): Cloud PAYG $0 ($0.02 per browser-hour, $5 minimum top-up; hosted agents add model cost + 20%). Status: verified. Source: https://browser-use.com/pricing
- Firecrawl (https://www.firecrawl.dev): Free $0 (1,000 credits/month); Hobby $19 (5,000 credits ($16/mo yearly)); Standard $99 (100,000 credits). Status: verified. Source: https://www.firecrawl.dev/pricing
- Exa (https://exa.ai): Pay as you go $0 (Search from $4 per 1,000 requests (Instant); $7 per 1,000 (Fast/Auto)). Status: verified. Source: https://exa.ai/pricing
- Tavily (https://www.tavily.com): Free $0 (1,000 credits/month); Pay as you go $0 ($0.008 per credit; basic search = 1 credit). Status: verified. Source: https://docs.tavily.com/documentation/api-credits
- Perplexity Pro (https://www.perplexity.ai/pro): Pro $17 (Price shown when billed yearly; monthly price not shown). Status: verified. Source: https://www.perplexity.ai/pro
- ChatGPT (https://chatgpt.com): Plus $20 (Monthly). Status: verified. Source: https://aiplantracker.com/compare/chatgpt-vs-perplexity/
- Fin by Intercom (https://fin.ai): Per outcome $0 ($0.99 per resolution, 50-outcome monthly minimum). Status: verified. Source: https://fin.ai/pricing
- Chatbase (https://www.chatbase.co): Hobby $40 (700 message credits ($32/mo yearly)); Standard $150 (4,000 credits); Pro $500 (15,000 credits). Status: verified. Source: https://www.chatbase.co/pricing
- n8n (https://n8n.io): Starter (€24/mo monthly (€20 yearly), 2,500 executions); Pro-1 (€60/mo monthly (€50 yearly), 10,000 executions). Status: verified. Source: https://n8n.io/pricing/
- Julius AI (https://julius.ai): Free $0; Plus $20 (2,000 credits ($16/mo yearly)); Pro $45 (5,000 credits). Status: verified. Source: https://julius.ai/pricing
- Vapi (https://vapi.ai): Usage $0 ($0.05/min platform fee plus STT, LLM, TTS and telephony at cost). Status: verified. Source: https://vapi.ai/pricing
- ElevenLabs Agents (https://elevenlabs.io/agents): Usage $0 ($0.08/min call minutes; LLM billed separately). Status: verified. Source: https://elevenlabs.io/pricing/agents
- Langfuse (https://langfuse.com): Hobby $0 (50k units/month, 2 users); Core $29 (100k units/month, unlimited users). Status: verified. Source: https://langfuse.com/pricing
- Mastra (https://mastra.ai): Open source $0 (Framework is free; pay for models). Status: verified. Source: https://dbggr.dev/articles/agent-frameworks-compared/
- LangGraph (https://www.langchain.com/langgraph): Open source $0 (Library is free; pay for models). Status: verified. Source: https://dbggr.dev/articles/agent-frameworks-compared/
- OpenAI Agents SDK (https://openai.github.io/openai-agents-python/): Open source $0 (Pay for model usage). Status: verified. Source: https://dbggr.dev/articles/agent-frameworks-compared/
- Claude Agent SDK (https://docs.claude.com/en/api/agent-sdk/overview): SDK $0 (Pay for Claude API usage). Status: verified. Source: https://dbggr.dev/articles/agent-frameworks-compared/
- discord.js (https://discord.js.org): Open source $0. Status: unverified. Source: https://discord.js.org
- Phaser (https://phaser.io): Open source $0. Status: unverified. Source: https://phaser.io
- MCP Inspector (https://github.com/modelcontextprotocol/inspector): Open source $0 (Run with npx). Status: unverified. Source: https://github.com/modelcontextprotocol/inspector
- Claude (app) (https://claude.com): Pro $20 ($17/mo billed yearly; includes Claude for Excel and Gmail/Calendar connectors); Max 5x $100 (5x Pro usage per session). Status: verified. Source: https://claude.com/pricing
- Lindy (https://www.lindy.ai): Free $0 ($50 in credits, valid 7 days, no card); Team $49.99 (Per active user, 5,000 credits/mo; top-ups $10 per 1,000 credits). Status: verified. Source: https://www.lindy.ai/pricing