@aiedx/firstbreakai

The official CLI and local MCP server for First Break AI — the first fully agentic AI cohort.

First Break AI is a cohort run by AI agents. AI agents answer your questions, validate your code against rubrics, and track your progress across the website, your terminal, and your IDE. The agentic backbone is powered by FetchLens.ai.

This package is how you put those AI agents in your own terminal and IDE: check your dev environment, track progress, validate exercises, scaffold a blog, and get AI-agent help — from the command line or any MCP-capable AI assistant (Cursor, Claude, Codex).

Install

npm install -g @aiedx/firstbreakai

Or run without installing:

npx @aiedx/firstbreakai doctor

Requires: Node.js >= 18

How AI agents run the cohort

First Break AI is completely agentic — every part of the cohort is powered by AI agents, not static videos and quizzes:

• AI agents answer your questions — the on-site assistant and firstbreakai ask know the syllabus, lessons, and your current step.
• AI agents validate your code — firstbreakai validate runs deterministic checks against the rubric and reports exactly what passed.
• AI agents track your progress — log in with Discord and your progress syncs across the CLI, the website, and the MCP server.
• AI agents live in your IDE — run firstbreakai mcp (or use the remote MCP server) so Cursor, Claude Desktop, Claude Code, and OpenAI Codex can call cohort tools directly.
• Powered by FetchLens.ai — the agentic AI platform that provides the MCP server backbone, the AI-agent assistant, and authenticated progress tracking.

Commands

| Command | Description | |---------|-------------| | firstbreakai doctor | Check your dev environment (Git, Python, Quarto, HF CLI, Node.js, Cursor) | | firstbreakai status | Show your cohort progress | | firstbreakai done <step> | Mark a step as complete | | firstbreakai init | Scaffold a new Quarto blog | | firstbreakai open <page> | Open a cohort page in your browser | | firstbreakai next | Open the next incomplete step | | firstbreakai ask "question" | Ask the FBA AI assistant | | firstbreakai login | Authenticate via Discord | | firstbreakai validate <step> | Run local validation checks for a step | | firstbreakai mcp | Start the MCP server for AI IDE integration |

Quick start

# 1. Check your environment
firstbreakai doctor

# 2. Scaffold a blog
mkdir my-ai-blog && cd my-ai-blog
firstbreakai init

# 3. Preview it
quarto preview

# 4. Mark step 1 as done
firstbreakai done 1

# 5. See what's next
firstbreakai next

MCP server — give your AI agent cohort superpowers

Run First Break AI as a local Model Context Protocol server so the AI agent inside your IDE (Cursor, Claude Desktop, Claude Code, OpenAI Codex) can call cohort tools directly — no context switching, fully agentic.

Add to your .cursor/mcp.json or Claude Desktop config:

{
  "mcpServers": {
    "firstbreakai": {
      "command": "npx",
      "args": ["@aiedx/firstbreakai", "mcp"]
    }
  }
}

This exposes these agentic cohort tools to your AI agent:

• cohort_doctor — environment check
• cohort_status — progress overview
• cohort_done — mark step complete
• cohort_validate — run validation checks
• cohort_open — open cohort pages
• cohort_next — open next step
• cohort_ask — ask the FBA AI assistant

Validation

firstbreakai validate <step> runs deterministic, local checks — no LLM, no network required.

Step 1 (Quarto blog): checks _quarto.yml, .qmd files, Git repo, GitHub remote.

Step 2 (Run a model locally): checks run.c/binary, config.json, tokenizer, model weights.

If you're logged in (firstbreakai login), results are also saved to your profile on the FBA server.

Links

• Cohort homepage
• Roadmap
• Lessons
• Checklist
• Discord
• GitHub

License

MIT