czlabs-code-mcp

Local MCP (Model Context Protocol) server that acts as a bridge to a RAG backend on Heroku.
Allows MCP-compatible editors/LLMs (e.g. Cursor) to use a RAG system over public repos (Jito/Agave) without exposing private keys.

🚀 Installation

npm i -g czlabs-code-mcp

⚙️ Cursor Configuration

Add to your configuration:

{
  "mcpServers": {
    "solana-code-mcp": {
      "command": "code-mcp",
      "args": [],
      "env": {
        "RAG_API_URL": "https://your-rag-app.herokuapp.com",
        "RAG_MCP_AUTH_TOKEN": "your-secret-token"
      }
    }
  }
}

• RAG_API_URL: URL de tu app Heroku (ej: https://your-rag-app.herokuapp.com)
• RAG_MCP_AUTH_TOKEN: your-secret-token

▶️ Usage

1. Restart Cursor.
2. Verify that the ping tool appears.
3. Execute ping → should respond pong.
4. Execute code_search or code_analysis → should call your Heroku API.

🔐 Security

• Your private keys (Qdrant, OpenAI) live only on Heroku.
• The local MCP never requires private keys.
• Use rotatable tokens to control access.

❓ FAQ

Do I need Heroku?
Yes, to expose your RAG API and keep private keys secure. The MCP is just the local bridge.

Does it work outside Cursor?
Yes, with any MCP client via stdio (other editors or apps that support MCP).

Why is there a ping tool?
To confirm that the MCP is working before using the RAG tools.

🛠️ Troubleshooting

• "No tools or prompts": check that there are no stdout logs and that the token is not validated in ListTools/CallTool.
• 401 errors on Heroku: check that RAG_MCP_AUTH_TOKEN is in your Cursor config.

🔗 Links

• MCP Documentation
• Cursor MCP Setup

Developed by CzLabs 🚀