MCP Proxy Hub is an MCP server that acts as a central gateway to aggregate and serve multiple downstream Model Context Protocol servers through a single unified interface. It connects to multiple backend MCP servers communicating over stdio, Server-Sent Events (SSE), or streamable HTTP transports, allowing developers, system architects, and AI agent operators to manage distributed tools and resources from a single endpoint. The hub routes tool calls, prompt requests, and resource resolution dynamically across registered backends while standardizing access through a JSON configuration file. It provides tool filtering, tool renaming, and compound custom tools that execute subtools across several servers in sequence. MCP Proxy Hub also handles sensitive variables by expanding environment variables in outgoing tool arguments and unexpanding sensitive values in server responses to prevent leaking secrets. Configurable transport settings allow the proxy itself to expose stdio, SSE, or streamable HTTP with optional bearer token authentication and custom request timeouts.
Category: AI & LLM Tooling
Tags: aggregation, configuration, gateway, proxy
bash git clone https://github.com/naotaka3/mcp-proxy-hub.git cd mcp-proxy-hub 2. Install dependencies and build the server: bash npm install npm run build 3. Create your configuration file from the template: bash cp config.example.json config.json 4. Add the proxy hub entry to your Claude Desktop configuration (~/Library/Application Support/Claude/claude_desktop_config.json on macOS or %APPDATA%/Claude/claude_desktop_config.json on Windows): json { "mcpServers": { "mcp-proxy-hub": { "command": "/path/to/mcp-proxy-hub/build/index.js", "env": { "MCP_PROXY_CONFIG_PATH": "/absolute/path/to/your/config.json", "KEEP_SERVER_OPEN": "1" } } } }Part of MCP Servers
MCP Proxy Hub is an aggregation gateway that connects to multiple Model Context Protocol servers across stdio, SSE, and streamable HTTP transports. It presents all connected tools, prompts, and resources through one unified interface, routing requests and combining responses based on a JSON configuration file.
MCP Proxy Hub works with any MCP-compliant client, including Claude Desktop and custom client implementations. It can expose its aggregated tools and resources over stdio, SSE, or streamable HTTP transports, allowing desktop applications and remote agent services to interact with it seamlessly.
The server includes an environment variable engine supporting both server-specific and global variables. You can mark variables to automatically expand using syntax like ${API_KEY} in tool arguments, and reverse-replace matching sensitive values back to variable references in responses to avoid leaking secrets.
Yes. Within the configuration file, you can specify exposedTools or hiddenTools for each downstream server. You can provide an array of tool names to restrict exposure, or pass objects with original and exposed properties to rename tools before exposing them to the client.
Yes. MCP Proxy Hub provides a dedicated CLI tool named mcp-proxy-hub-cli. It supports direct execution commands to list and call tools with argument inputs, as well as an interactive shell session for executing commands and saving responses directly to disk.