pragun-clique-labs/mcp-executor

MCP Executor

A FastAPI server that creates and executes MCP (Model Context Protocol) tool bundles.

Features

• Bundle Creation: Break down descriptions into executable tasks using available MCP tools
• Route Execution: Execute predefined routes or create new ones on-the-fly
• LangChain Integration: Uses LangChain for task planning and execution
• Supabase Storage: Persists bundles and routes in Supabase database

Endpoints

POST /mcp-bundler

Creates a bundle by analyzing MCP tools and breaking down a description into executable routes.

Request:

{
  "project_id": "uuid",
  "mcp_ids": ["uuid1", "uuid2"],
  "description": "Description of what you want to accomplish"
}

Response:

{
  "bundle_id": "uuid",
  "routes_created": 3
}

POST /executor

Executes a request using existing routes or creates new routes on-the-fly.

Request:

{
  "bundle_id": "uuid",
  "request": "What you want to do"
}

Response:

{
  "result": "Execution results",
  "route_used": "route_uuid",
  "new_route_created": false
}

Setup

1. Install dependencies:

pip install -r requirements.txt

2. Set environment variables:

export SUPABASE_URL=your_supabase_url
export SUPABASE_ANON_KEY=your_supabase_anon_key
export ANTHROPIC_API_KEY=your_anthropic_api_key

3. Run the server:

python main.py

How it Works

1. Bundle Creation: The bundler endpoint analyzes available MCP tools and uses LangChain to break down the description into specific tasks, creating routes for each task.

2. Execution: The executor endpoint finds matching routes for requests or creates new routes dynamically, then executes them using the MCP tools through LangChain agents.

3. Storage: All bundles and routes are stored in Supabase for reuse and tracking.