MCP Discovery

The World's Largest MCP Server Index - 14,000+ Servers

Live API: https://mcp-discovery-two.vercel.app

LangChain Integration - NEW

Enable your LangChain agents to dynamically discover MCP servers based on task requirements.

from langchain_mcp_discovery import MCPDiscoveryTool

# Create the discovery tool
discovery = MCPDiscoveryTool()

# Agent discovers tools dynamically
result = discovery.run("I need to query a PostgreSQL database")
# Returns: Server name, install command, performance metrics

Features:

• Semantic search across 14,000+ MCP servers
• Performance metrics (latency, uptime)
• Ready-to-use install commands
• Full LangChain BaseTool compatibility

See langchain/ for full documentation, examples, and tests.

Not for Humans. Built for Agents.

MCP Discovery is NOT a website for humans to browse. It's a machine-to-machine API designed for AI agents to programmatically discover tools.

With 14,000+ MCP servers indexed, no human could reasonably browse this directory. But an AI agent can query it in milliseconds and find exactly what it needs.

Human: "Help me analyze my GitHub repos and save to a database"

Agent thinks: "I need GitHub access and database storage"

Agent calls: POST /api/v1/discover {"need": "github api access"}
Agent calls: POST /api/v1/discover {"need": "database storage"}

Agent receives: github-server, sqlite-server, postgres-server
Agent installs tools and completes the task autonomously

The agent never asks the human which MCP server to use. It just figures it out.

Why 14,000+ Servers Matters

Scale matters. The more servers indexed, the more capable your agent becomes.

Quick Start (For Agent Developers)

Semantic Search - Find Any Tool

curl -X POST https://mcp-discovery-two.vercel.app/api/v1/discover \
  -H "Content-Type: application/json" \
  -d '{"need": "send slack notifications"}'

Response:

{
  "recommendations": [
    {
      "server": "slack-server",
      "name": "Slack MCP Server",
      "install_command": "npx -y @anthropic/mcp-server-slack",
      "confidence": 0.72,
      "description": "Slack integration for MCP...",
      "category": "communication"
    }
  ],
  "total_found": 5,
  "query_time_ms": 48
}

List All Categories

curl https://mcp-discovery-two.vercel.app/api/v1/categories

Browse by Category

curl "https://mcp-discovery-two.vercel.app/api/v1/servers?category=database&limit=50"

Index Statistics

14,000+ MCP Servers from All Major Sources

Categories (21+)

Categories are auto-assigned. "Other" contains specialized tools.

API Reference

POST /api/v1/discover

The primary endpoint for agents. Accepts natural language queries.

Request:

{
  "need": "I need to read and write files on the local filesystem",
  "limit": 10
}

Response:

{
  "recommendations": [
    {
      "server": "filesystem-server",
      "name": "Filesystem Server",
      "npm_package": "@modelcontextprotocol/server-filesystem",
      "install_command": "npx -y @modelcontextprotocol/server-filesystem",
      "confidence": 0.85,
      "description": "Secure file operations for MCP...",
      "category": "development",
      "github_url": "https://github.com/modelcontextprotocol/servers"
    }
  ],
  "total_found": 10,
  "query_time_ms": 52
}

Integration Examples

Autonomous Agent (Python)

import requests

class AutonomousAgent:
    def discover_tools(self, task_description: str) -> list:
        """Agent discovers needed tools without human intervention"""
        response = requests.post(
            "https://mcp-discovery-two.vercel.app/api/v1/discover",
            json={"need": task_description, "limit": 5}
        )
        return response.json()["recommendations"]

    def execute_task(self, user_request: str):
        # Agent analyzes request and discovers tools
        tools = self.discover_tools(user_request)

        # Agent installs first recommended tool
        install_cmd = tools[0]["install_command"]
        # ... agent proceeds autonomously

AI Coding Assistant (JavaScript)

// Agent middleware for Cursor/Continue/Windsurf
async function discoverMCPServer(need) {
  const response = await fetch('https://mcp-discovery-two.vercel.app/api/v1/discover', {
    method: 'POST',
    headers: { 'Content-Type': 'application/json' },
    body: JSON.stringify({ need, limit: 3 })
  });
  return response.json();
}

// Usage in agent loop
const tools = await discoverMCPServer('database for storing user preferences');
// Returns: sqlite-server, redis-server, postgres-server

Claude Desktop Integration

Add discovery as an MCP server itself:

{
  "mcpServers": {
    "discovery": {
      "command": "npx",
      "args": ["-y", "mcp-discovery-api"]
    }
  }
}

Performance

Tech Stack

• Database: Supabase (PostgreSQL + pgvector for semantic search)
• Embeddings: OpenAI text-embedding-3-small (1536 dimensions)
• API: Vercel Serverless Functions (Edge-optimized)
• Scraping: Direct API integration + DeepSeek for extraction

Self-Host

Environment Variables

SUPABASE_URL=your-supabase-url
SUPABASE_SERVICE_ROLE_KEY=your-service-role-key
OPENAI_API_KEY=your-openai-api-key

Deploy

git clone https://github.com/yksanjo/mcp-discovery.git
cd mcp-discovery
npm install
cp .env.example .env
# Configure your credentials
vercel --prod

The Vision

As AI agents become more autonomous, they need to discover their own tools. MCP Discovery is the infrastructure layer that enables this.

Today: 14,000+ servers indexed Goal: Every MCP server ever created, instantly discoverable

Links

• Live API: https://mcp-discovery-two.vercel.app
• Health Check: https://mcp-discovery-two.vercel.app/health
• Categories: https://mcp-discovery-two.vercel.app/api/v1/categories
• Model Context Protocol
• Anthropic MCP Servers

License

MIT

14,000+ MCP Servers. One API. Built for Machines.