Generic MCP Search-Fetch Server

A generic MCP server that provides OpenAI-compliant search and fetch tools without any external API dependencies. This server can be easily customized to work with any data source.

Features

• OpenAI Compliant: Implements the exact tool specifications required for ChatGPT connectors and deep research
• No External Dependencies: Works with local data sources without requiring API keys
• Easy Customization: Simple to adapt for different data backends (database, API, vector store, etc.)
• TypeScript Implementation: Modern, type-safe codebase

Quick Start

# Install dependencies
npm install

# Development mode
npm run dev

# Build and run
npm run build
npm start

Run the Published Package

Recent npm releases (10.x/11.x) ship npx as an alias of npm exec. In this mode the classic npx <pkg>@<version> syntax no longer injects the package's bin into PATH, so the shell prints command not found. Use the explicit --package flag instead:

npx --yes --package [email protected] generic-mcp-search-fetch --stdio

To always track the latest publish, swap @1.0.6 for @latest. The command keeps the process alive and speaks MCP over STDIO.

MetaMCP Configuration

When adding the server in MetaMCP (STDIO transport):

• Command: /opt/homebrew/bin/npx (or your system's npx path)
• Arguments: --yes --package [email protected] generic-mcp-search-fetch --stdio

As an alternative that bypasses npx, resolve the published CLI and run it with Node:

node $(node -e "const p=require.resolve('generic-mcp-search-fetch/package.json'); const {dirname,join}=require('path'); console.log(join(dirname(p),'dist/cli.js'));") --stdio

Cline Configuration

Add the server to cline_mcp_settings.json (macOS path: ~/Library/Application Support/Code/User/globalStorage/saoudrizwan.claude-dev/settings/cline_mcp_settings.json):

{
  "mcpServers": {
    "search_fetch": {
      "timeout": 60,
      "type": "stdio",
      "command": "npx",
      "args": [
        "--yes",
        "--package",
        "[email protected]",
        "generic-mcp-search-fetch",
        "--stdio"
      ]
    }
  }
}

Customize the server key (search_fetch) or add autoApprove/env blocks as needed for your workflow.

ChatGPT Connector Setup

When adding this MCP server in ChatGPT → Settings → Connectors:

• Command: node
• Arguments: dist/server.js (or npm run dev for development)

Tools

search Tool

Searches through the data corpus and returns matching results.

Input:

{
  query: string,      // Search query
  limit?: number,     // Maximum results (default: 10, max: 50)
  cursor?: string     // Pagination cursor (optional)
}

Output:

{
  results: [{
    id: string,       // Unique identifier for fetching
    title: string,    // Human-readable title
    snippet?: string, // Preview snippet
    url?: string      // Source URL for citation
  }],
  nextCursor?: string | null // For pagination
}

fetch Tool

Retrieves complete document content by ID.

Input:

{
  id: string  // Document ID from search results
}

Output:

{
  id: string,           // Document ID
  title?: string,       // Document title
  content: string,      // Full document content
  url?: string,         // Source URL
  metadata?: object     // Additional metadata
}

Customization

Data Source

Replace the data in data/corpus.json with your own content. The file structure should be:

[
  {
    "id": "unique_id",
    "title": "Document Title",
    "url": "https://example.com/source",
    "content": "Full document content here..."
  }
]

Backend Integration

To connect to a different data source (database, API, etc.), modify the loadCorpus() function in src/server.ts and the search logic to match your backend.

Development

# Development with hot reload
npm run dev

# Build for production
npm run build

# Run production build
npm start

License

MIT License - Feel free to customize for your specific use case.