cofound-mcp

MCP server that makes your AI coding assistant business-aware.

Plug it into Claude Code, Cursor, or Windsurf. From then on your AI knows your avatar, grades your copy, scores your offer, and builds your funnel — all using Alex Hormozi's $100M frameworks as structured data.

Free until 1,000 daily active users. No API key. No account.

Install (developers)

npm install -g cofound-mcp

Requires Node.js 18 or higher. To check: node --version

If you don't have Node.js: nodejs.org/download — download the LTS version, run the installer, then come back here.

Set up your business

Run this once in your project folder:

cofound init

It walks you through:

1. Who you're selling to (avatar)
2. What they want (dream outcome)
3. What's in your offer
4. What your funnel looks like

Takes about 20 minutes. Writes cofound.config.json at your project root. Every Claude session after this knows your business.

Connect to your editor

Claude Code

Create .mcp.json at your project root:

{
  "mcpServers": {
    "cofound": {
      "command": "cofound-mcp"
    }
  }
}

Cursor

Create .cursor/mcp.json:

{
  "mcpServers": {
    "cofound": {
      "command": "cofound-mcp"
    }
  }
}

Windsurf

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "cofound": {
      "command": "cofound-mcp"
    }
  }
}

Tools

| Tool | What it does | |------|-------------| | cofound_init | Set up your avatar, offer, and funnel | | cofound_status | See what's configured and what's missing | | cofound_analyze_copy | Grade copy on reading level, pain, and avatar fit | | cofound_rewrite_copy | Rewrite copy to pass grade 3 reading level and pain tests | | cofound_audit_codebase | Scan every .copy.json and rank by how badly they're failing | | cofound_scaffold_copy | Pull hardcoded strings out of a component into a .copy.json file | | cofound_check_reading_level | Quick reading level check (Flesch-Kincaid, Coleman-Liau, ARI) | | cofound_validate_headline | Score headline on the MAGIC formula, get 2–3 stronger variants | | cofound_grade_offer | Score your offer on the Value Equation | | cofound_suggest_guarantee | Get the right guarantee for your price point + script to use it | | cofound_recommend_funnel | Recommend a funnel type based on price and traffic temperature | | cofound_funnel_report | Show actual vs benchmark conversions at each funnel stage | | cofound_track | Record a funnel event (used by the tracking pixel) |

Example usage

Ask Claude Code (or Cursor, or Windsurf):

Grade the copy on my homepage.

Rewrite my hero headline. Hit grade 3 reading level.

Score my offer against the Value Equation. Tell me what's weak.

Audit every .copy.json in this project and tell me what to fix first.

The .copy.json pattern

cofound works best when each component has a .copy.json file next to it:

src/
  pages/
    home/
      home.tsx
      home.copy.json   ← all the words

home.copy.json:

{
  "meta": {
    "component": "home",
    "route": "/",
    "avatar": "indie_builder",
    "last_graded": "2026-03-08",
    "reading_level": 3.0,
    "pain_score": 7,
    "slop_score": 2
  },
  "copy": {
    "headline": "Your AI writes code. Your site still doesn't sell.",
    "subheadline": "cofound fixes that.",
    "cta": "Get It Free"
  }
}

Your component imports from the copy file. cofound scans all copy files, grades them, and rewrites them. The component code never changes — only the words.

Don't have this pattern yet? Run:

cofound_scaffold_copy src/pages/home/home.tsx

It extracts the strings and writes the file for you.

Opt out of telemetry

cofound counts anonymous install events and tool runs. No personal data, no file contents, no config values — ever. The exact payload is in src/telemetry.ts.

To opt out:

COFOUND_NO_TELEMETRY=1 cofound-mcp

Or set it permanently in your shell profile:

echo 'export COFOUND_NO_TELEMETRY=1' >> ~/.zshrc

License

MIT