Carrot AI PM is an open-source MCP server that lives inside Cursor and closes the gap between “prompt” and production-ready code.

• Generates structured, enforceable specs (OpenAPI, UI, DB, CLI)

• Runs AST-level checks to verify the AI implementation matches the spec

• Flags skipped validation, missing auth, hallucinated functions, etc.

• Stores every result (.carrot/compliance/*) for audit trails and trend tracking

• 100 % local – Carrot never calls external APIs; it just uses Cursor’s internal LLM hook

Quick start: git clone https://github.com/talvinder/carrot-ai-pm npm install && npm run build # add to .cursor/mcp.json, then chat: # “Create spec for X → implement → check implementation”

Example output: endpoints match missing rate-limit on POST /login undefined fn: sendWelcomeEmail()

I built it after years of watching markdown specs drift. Edge-cases, bugs, and PRs welcome → https://github.com/talvinder/carrot-ai-pm/issues