mcptool

Use 100+ MCP tools without blowing your context window.

A drop-in MCP proxy that sits between your AI client (Claude Desktop, Cursor, VS Code) and multiple backend MCP servers. Instead of exposing all tool definitions (100+ tools, 50K+ tokens), MCPtool exposes just 2 meta-tools:

• search_tools(query) — find relevant tools by keyword (BM25 search)
• call_tool(name, arguments) — invoke the tool on the correct backend server

Why

Without MCPtool	With MCPtool
100+ tools in every request	2 tools always
50,000+ tokens per call	~500 tokens per call
Context window blown	Context window intact

Install

npx mcptool --config mcptool.json

Or install globally:

npm install -g mcptool
mcptool --config mcptool.json

Configuration

Create a mcptool.json file:

{
  "servers": [
    {
      "name": "github",
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": { "GITHUB_TOKEN": "your-token-here" }
    },
    {
      "name": "filesystem",
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/home/user/docs"]
    }
  ]
}

Claude Desktop Setup

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "mcptool": {
      "command": "npx",
      "args": ["mcptool", "--config", "/path/to/mcptool.json"]
    }
  }
}

How it works

1. Startup: MCPtool connects to all configured backend MCP servers and calls tools/list on each
2. Indexing: All tool names + descriptions are indexed with BM25 for keyword search
3. Runtime: Your AI client sees only search_tools and call_tool
4. Discovery: LLM calls search_tools("create github issue") → gets back relevant tools
5. Execution: LLM calls call_tool("create_issue", {...}) → MCPtool routes to the right backend

Development

git clone https://github.com/kspatel29/mcptool
cd mcptool
npm install
npm run build
npm test

License

MIT