MCPFinder

MCPfinder 🔧🤖 (@mcpfinder/server)

Meet the simplest way to supercharge your coding and AI agents with MCP — an "API for AI." MCPfinder enables language models to search for and install new capabilities on demand through client applications that support the MCP protocol. No coding or manual setup required.

App Store for Agents

Users can request tools the AI doesn't have yet, or the AI can autonomously expand its own capabilities by discovering relevant MCP servers. From code generators and data analyzers to specialized knowledge tools, MCPfinder acts like a map and toolbox for AI — transforming static models into evolving, capability-hunting agents that grow more powerful with every interaction.

Plug-and-Play Tools for LLMs

This Node.js application implements an MCP (Model Context Protocol) server designed to run locally alongside MCP clients (like Cursor, Claude Desktop, Windsurf). It provides tools that allow AI within the client to interact with the MCP ecosystem.

Specifically, it enables the AI assistant to:

1. Discover available MCP servers registered in the central MCPfinder Registry (via search_mcp_servers).
2. Retrieve details about specific servers (via get_mcp_server_details).
3. Manage the client application's local MCP server configuration file (add/update via add_mcp_server_config, remove via remove_mcp_server_config).

Note for AI Agents: This server is the primary interface for finding and enabling new tools and capabilities requested by the user if they aren't already available in current MCP toolset. Use search_mcp_servers first when asked by the user for a capability you don't possess.

⚡ Instant Availability

We curate only MCP servers that are instantly available to AI agents — no manual setup, no cloning repositories, no complex installation steps. Just one command or API call, and your AI gains new capabilities immediately.

Zero-Friction Setup

All servers work via npx (Node.js packages), uvx (Python packages), or direct HTTP/SSE endpoints

AI-Ready

Your AI can discover and install tools autonomously, no human intervention needed

Real-Time Enhancement

Capabilities expand on-demand, exactly when your conversation needs them

Quick Start

Option 1: Local Installation (stdio transport)

Run in your terminal the interactive setup tool to automatically update the MCP configuration file:

npx -y @mcpfinder/server install

This command guides you through selecting your client (Cursor, VS Code, Claude, etc.) and adds the necessary mcpfinder entry to the correct configuration file (e.g., ~/.cursor/mcp.json).

Option 2: Cloud Access (HTTP/SSE transport)

You can also access MCPfinder directly via HTTP/SSE transport without local installation:

# For Claude CLI users
claude mcp add --transport http mcpfinder https://mcpfinder.dev/mcp

# Direct HTTP endpoint
https://mcpfinder.dev/mcp

This provides the same tools (search_mcp_servers, get_mcp_server_details, etc.) but runs in the cloud without needing local Node.js or npm.

See "Running from source" and "Commands and Options" for more details if you are working directly with the source code.

Manual Configuration

To manually configure an MCP client, you need to create or modify its JSON configuration file to include an entry for mcpfinder.

Stdio Transport Configuration (Local)

Configuration File Structure:

{
  "mcpServers": { 
    "mcpfinder": {
      "command": "npx",
      "args": [
        "-y",
        "@mcpfinder/server"
      ]
    },
  }
}

Note: For Visual Studio Code (settings.json), the top-level key for MCP configurations must be servers instead of mcpServers.

HTTP/SSE Transport Configuration (Cloud)

For clients that support HTTP/SSE transport:

{
  "mcpServers": {
    "mcpfinder": {
      "url": "https://mcpfinder.dev/mcp",
      "transport": "http"
    }
  }
}

Running from source

• Clone this repository, e.g., git clone https://github.com/mcpfinder/server
• Run node index.js for Stdio mode or node index.js --http for HTTP mode.

Commands and Options

When running from source (node index.js), the script can be invoked in several ways:

Running the Server (Default Behavior): If no command is specified, index.js starts the MCP server.

• Stdio Mode (default):

  node index.js
  

• HTTP Mode (local):

  node index.js --http
  

  • --port <number>: Specify the port for HTTP mode (default: 6181, or MCP_PORT env var).
  • --api-url <url>: Specify the MCPfinder Registry API URL used by the tools (default: https://mcpfinder.dev, or MCPFINDER_API_URL env var).

Note: The HTTP mode runs locally on your machine. For cloud-based HTTP/SSE access, use the public endpoint at https://mcpfinder.dev/mcp instead.

Executing Commands:

• install: Run the interactive setup to configure a client application.

  node index.js install
  

• register: For server publishers to register their MCP server package with the MCPFinder registry.

  node index.js register [package-name-or-url] [options]
  # or when installed globally:
  npx -y @mcpfinder/server register [package-name-or-url] [options]
  

  Register Command Options:

  • --headless: Run registration without interactive prompts (uses defaults)
  • --use-uvx: Specify that the package should be run with uvx instead of npx (for Python packages)
  • --description <text>: Provide a description for the server
  • --tags <tags>: Comma-separated list of tags (e.g., "database,api,search")
  • --auth-token <token>: Authentication token for the registry
  • --requires-api-key: Indicate that the server requires an API key
  • --auth-type <type>: Type of authentication (default: "api-key")
  • --key-name <name>: Name of the API key environment variable
  • --auth-instructions <text>: Instructions for obtaining API keys
  • --confirm <y/n>: Auto-confirm registration without manual approval
  • --manual-capabilities <y/n>: Manually specify capabilities instead of auto-detection
  • --has-tools <y/n>: Specify if the server provides tools
  • --has-resources <y/n>: Specify if the server provides resources
  • --has-prompts <y/n>: Specify if the server provides prompts

  This command will:

  • Accept npm package names (e.g., @username/my-mcp-server) or HTTP/SSE endpoints
  • Support both Node.js packages (npx) and Python packages (uvx)
  • Automatically connect to your MCP server to verify it's valid
  • Introspect available tools, resources, and prompts
  • Generate a manifest with your server's capabilities
  • Submit it to the MCPfinder registry

Getting Help:

• --help: Display the help message detailing commands and options.

  node index.js --help
  

The server uses the following environment variables:

• MCPFINDER_API_URL: The base URL for the MCPfinder Registry API. Defaults to https://mcpfinder.dev.
• MCP_PORT (HTTP Mode Only): The port number for the server to listen on. Defaults to 6181.

Provided Tools

This MCP server exposes the following tools to the connected AI assistant (available via both stdio and HTTP/SSE transports):

1. search_mcp_servers

• Description: Searches the MCPfinder Registry for available MCP servers. This is the primary tool for discovering and accessing new tools, methods, features, or capabilities.
• Input Schema:
  • query (string, optional): Keywords to search for in tool name or description.
  • tag (string, optional): Specific tag to filter by.
• Output: A list of matching server summaries (server_id, name, description, URL, tags). The typical next step is to use get_mcp_server_details for more info or directly add_mcp_server_config to install one.

⚠️ Note: The registry currently contains several hundred servers that can be run locally using npx (Node.js) or uvx (Python) in stdio mode without requiring environment variables for basic operation. Future updates will expand support to include a wider range of servers, including paid and commercial options that require environment keys.

2. get_mcp_server_details

• Description: Retrieves detailed information about a specific MCP server from the registry, including its full manifest and basic installation suggestions (command, environment variables). Use this after finding a server_id via search_mcp_servers to get more information before potentially adding it.
• Input Schema:
  • id (string, required): The unique MCPfinder's server_id obtained from search_mcp_servers.
• Output: The detailed server manifest and installation hints. The next step is to use add_mcp_server_config to install the server.

3. add_mcp_server_config

• Description: Adds or updates the configuration for a specific MCP server in the client application's local configuration file (e.g., Cursor's ~/.cursor/mcp.json). You must provide either client_type OR config_file_path.
• Input Schema:
  • server_id (string, required): A unique identifier for the server configuration entry (the MCPfinder ID obtained from search_mcp_servers).
  • client_type (string, optional): The type of client application (known types determined dynamically, examples: 'cursor', 'claude', 'windsurf'). Mutually exclusive with config_file_path. Use this for standard client installations.
  • config_file_path (string, optional): An absolute path or a path starting with ~ (home directory) to the target JSON configuration file (e.g., /path/to/custom/mcp.json or ~/custom/mcp.json). Mutually exclusive with client_type. Use this for non-standard locations or custom clients.

---

README truncated. View full README on GitHub.