metamcp: Combine MCP Servers Behind One Gateway

Overview

MetaMCP is a self-hostable MCP gateway that aggregates multiple MCP servers into a single server endpoint. Instead of configuring each upstream server separately in every MCP client, operators can group servers into namespaces, choose the tools to expose, and connect clients to a unified endpoint.

It is most useful as shared infrastructure for teams building agents or managing MCP integrations. It also provides authentication and endpoint management for deployments that need more control than a local collection of server configurations.

Key Features

Aggregate tools, resources, and prompts from multiple MCP servers.
Group servers into namespaces, with controls to enable or disable servers and individual tools.
Expose namespaces through SSE or Streamable HTTP MCP endpoints, as well as OpenAPI endpoints.
Secure endpoints with API keys or MCP OAuth, and support OIDC sign-in for the management interface.
Rename tools, adjust their descriptions, and attach namespace-specific annotations.
Apply namespace-level middleware, including a built-in option to filter inactive tools.
Inspect endpoints and save server configurations for debugging.
Configure endpoint-wide and per-user rate limits.

Use Cases

Agent developers can give an agent one endpoint for tools supplied by several MCP servers.
Platform teams can centrally host MCP integrations and manage which tools are available in each namespace.
Organizations with multiple users can use authentication, registration controls, and private or public access scopes to manage access.
Open WebUI or other compatible clients can use the project's OpenAPI endpoints when an MCP connection is not the desired interface.

Project Facts

Language: TypeScript
License: MIT
Stars: 2.7k
Forks: 436
Topics: mcp, mcp-gateway, mcp-host, mcp-middleware, mcp-server, mcp-servers, mcp-to-openapi, mcp-tools, model-context-protocol, model-context-protocol-server, model-context-protocol-servers, open-webui, openapi, self-hosted, vibe-coding
Archived: No

Getting Started

The recommended deployment is Docker Compose:

git clone https://github.com/metatool-ai/metamcp.git
cd metamcp
cp example.env .env
docker compose up -d

See the README and documentation for configuration and client connection details.

Alternatives

MCPJungle: MCPJungle focuses on registering and sharing MCP servers, while MetaMCP emphasizes unified endpoints with namespaces and middleware.
mcp-gateway: mcp-gateway also fronts REST APIs and uses on-demand tool discovery through a compact meta-tool, unlike MetaMCP's configured server aggregation.
mcphub: MCPHub emphasizes centralized credentials, logs, and health monitoring, while MetaMCP highlights namespaces and middleware for unified MCP endpoints.

Considerations

This is a server deployment, not just a client-side MCP configuration tool. The README recommends at least 2–4 GB of memory for online or VPS deployments.
MCP servers that require dependencies beyond uvx or npx may need a customized Dockerfile.
The project notes that the ai-dev branch is the forward development branch and advises testing before building an image from it.
Rate-limit counters are in memory, and each machine in a cluster tracks only its own traffic.
SSE connections may require appropriate reverse-proxy configuration. The README includes an Nginx example.