An MCP proxy server that aggregates and serves multiple MCP resource servers through a single HTTP server.

• Proxy Multiple MCP Clients: Connects to multiple MCP resource servers and aggregates their tools and capabilities.
• SSE Support: Provides an SSE (Server-Sent Events) server for real-time updates.
• Flexible Configuration: Supports multiple client types (stdio, sse or streamable-http) with customizable settings.

git clone https://github.com/TBXark/mcp-proxy.git
cd mcp-proxy
make build
./build/mcp-proxy --config path/to/config.json

go install github.com/TBXark/mcp-proxy@latest

The Docker image supports two MCP calling methods by default: npx and uvx.

docker run -d -p 9090:9090 -v /path/to/config.json:/config/config.json ghcr.io/tbxark/mcp-proxy:latest
# or 
docker run -d -p 9090:9090 ghcr.io/tbxark/mcp-proxy:latest --config https://example.com/path/to/config.json

The server is configured using a JSON file. Below is an example configuration:

This is the format for the new version's configuration. The old version's configuration will be automatically converted to the new format's configuration when it is loaded.

You can use https://tbxark.github.io/mcp-proxy to convert the configuration of mcp-proxy into the configuration that Claude can use.

{
  "mcpProxy": {
    "baseURL": "https://mcp.example.com",
    "addr": ":9090",
    "name": "MCP Proxy",
    "version": "1.0.0",
    "options": {
      "panicIfInvalid": false,
      "logEnabled": true,
      "authTokens": [
        "DefaultTokens"
      ]
    }
  },
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "<YOUR_TOKEN>"
      },
      "options": {
        "toolFilter": {
          "mode": "block",
          "list": [
            "create_or_update_file"
          ]
        }
      }
    },
    "fetch": {
      "command": "uvx",
      "args": [
        "mcp-server-fetch"
      ],
      "options": {
        "panicIfInvalid": true,
        "logEnabled": false,
        "authTokens": [
          "SpecificTokens"
        ]
      }
    },
    "amap": {
      "url": "https://mcp.amap.com/sse?key=<YOUR_TOKEN>"
    }
  }
}

Common options for mcpProxy and mcpServers.

• panicIfInvalid: If true, the server will panic if the client is invalid.
• logEnabled: If true, the server will log the client's requests.
• authTokens: A list of authentication tokens for the client. The Authorization header will be checked against this list.
• toolFilter: Optional tool filtering configuration. This configuration is only effective in mcpServers.
  • mode: Specifies the filtering mode. Must be explicitly set to allow or block if list is provided. If list is present but mode is missing or invalid, the filter will be ignored for this server.
  • list: A list of tool names to filter (either allow or block based on the mode).

  Tip: If you don't know the exact tool names, run the proxy once without any toolFilter configured. The console will log messages like <server_name> Adding tool <tool_name> for each successfully registered tool. You can use these logged names in your toolFilter list.

In the new configuration, the authTokens of mcpProxy is not a global authentication token, but rather the default authentication token for mcpProxy. When authTokens is set in mcpServers, the value of authTokens in mcpServers will be used instead of the value in mcpProxy. In other words, the authTokens of mcpProxy serves as a default value and is only applied when authTokens is not set in mcpServers.

Other fields are the same.

Proxy HTTP server configuration

• baseURL: The public accessible URL of the server. This is used to generate the URLs for the clients.
• addr: The address the server listens on.
• name: The name of the server.
• version: The version of the server.
• options: Default options for the mcpServers.

MCP server configuration, Adopt the same configuration format as other MCP Clients.

• transportType: The transport type of the MCP client. Except for streamable-http, which requires manual configuration, the rest will be automatically configured according to the content of the configuration file.
  • stdio: The MCP client is a command line tool that is run in a subprocess.
  • sse: The MCP client is a server that supports SSE (Server-Sent Events).
  • streamable-http: The MCP client is a server that supports HTTP streaming.

For stdio mcp servers, the command field is required.

• command: The command to run the MCP client.
• args: The arguments to pass to the command.
• env: The environment variables to set for the command.
• options: Options specific to the client.

For sse mcp servers, the url field is required. When the current url exists, sse will be automatically configured.

• url: The URL of the MCP client.
• headers: The headers to send with the request to the MCP client.

For http streaming mcp servers, the url field is required. and transportType need to manually set to streamable-http.

• url: The URL of the MCP client.
• headers: The headers to send with the request to the MCP client.
• timeout: The timeout for the request to the MCP client.

Usage of mcp-proxy:
  -config string
        path to config file or a http(s) url (default "config.json")
  -help
        print help and exit
  -version
        print version and exit

1. The server will start and aggregate the tools and capabilities of the configured MCP clients.
2. You can access the server at http(s)://{baseURL}/{clientName}/sse. (e.g., https://mcp.example.com/fetch/sse, based on the example configuration)
3. If your MCP client does not support custom request headers., you can change the key in clients such as fetch to fetch/{authToken}, and then access it via fetch/{authToken}.

• This project was inspired by the adamwattis/mcp-proxy-server project
• If you have any questions about deployment, you can refer to 《在 Docker 沙箱中运行 MCP Server》(@ccbikai)

This project is licensed under the MIT License. See the LICENSE file for details.