Tool Gating MCP

Name: Tool Gating MCP
Author: ajbmachon

@ajbmachon

代码仓库

关于 Tool Gating MCP

Implements a tool gating and discovery system to preserve LLM context and focus while having access to many MCP servers

#tool-gating #mcp #proxy

基本信息

分类

其他

运行时

python

传输方式

stdio

发布者

ajbmachon

配置

使用下面的配置,将此服务器添加到你的 MCP 客户端。

{
  "mcpServers": {
    "tool-gating-mcp": {
      "command": "uv",
      "args": [
        "venv"
      ]
    }
  }
}

工具

未检测到工具

工具是从 README 中自动提取的。维护者可以在 ## Tools 标题下列出工具,即可填充这部分内容。

概览

What is Tool Gating MCP?

Tool Gating MCP is an intelligent proxy/router for the Model Context Protocol (MCP) that allows clients like Claude Desktop to dynamically discover and use tools from multiple MCP servers through a single connection point. It prevents context bloat by selecting only the most relevant tools for each task.

How to use Tool Gating MCP?

Install Python 3.12+ and uv, clone the repo, create a virtual environment, install dependencies, and run tool-gating-mcp (HTTP API server) or integrate it as a native MCP server using mcp-proxy. Configure Claude Desktop to point to Tool Gating MCP; then use built-in tools like discover_tools, provision_tools, and register_tool.

Key features of Tool Gating MCP

Proxy architecture routes calls to multiple backend MCP servers
Dynamic tool discovery via semantic search across all servers
Smart provisioning loads only relevant tools within token budgets
Transparent execution routes tool calls to the correct backend
90%+ reduction in context usage compared to loading all servers
Zero configuration – Claude Desktop only needs Tool Gating MCP

Use cases of Tool Gating MCP

Reducing context bloat when using many MCP servers (e.g., Exa, Puppeteer, Context7, Desktop Commander)
Dynamically adding tools during a conversation without restarting Claude Desktop
Improving AI tool selection by presenting only the most relevant tools per query
Lowering token costs by minimizing the number of tool definitions sent per request

FAQ from Tool Gating MCP

What problem does Tool Gating MCP solve?

It solves context bloat caused by loading 100+ tools from multiple MCP servers at startup, and removes the need to restart Claude Desktop to add new servers during a conversation.

How does it reduce context usage?

It uses semantic search and token budget enforcement to select and provision only the most relevant tools (typically 2–3) out of a large pool, cutting token consumption by over 90%.

What are the prerequisites?

Python 3.12+, the uv package manager, and optionally mcp-proxy for native MCP integration with Claude Desktop.

How does tool discovery work?

Tools are registered with metadata, tags, and token estimates. User queries are embedded using a sentence transformer, and tools are scored by cosine similarity plus tag matches (0.2 boost per matching tag).

Can I integrate new MCP servers with Tool Gating MCP?

Yes. You can register tools from external MCP servers via the API endpoints (/api/tools/register) or use the AI-assisted server registration endpoint (/api/mcp/ai/register-server).