mcphub: Connect and Manage MCP Servers Through One Gateway

Summary
MCPHub is a self-hosted gateway and control plane for connecting MCP clients to local and remote MCP servers. It centralizes routing, access controls, credentials, logs, and health monitoring.
At a glance
- Language
- TypeScript
- License
- Apache-2.0
- Stars
- 2.5k
- Forks
- 330
- Added to OSRepos
- October 21, 2025
- Last analyzed
- October 3, 2026
Topics
Click on any tag to explore related repositories
Use at your own risk
OSRepos shares public repositories for knowledge and discovery only. Any installation, execution, configuration, or use of code from these repositories is the user's own responsibility. Always review the repository, source code, dependencies, licenses, and security implications before running or installing anything. OSRepos is not responsible for issues, damages, or losses resulting from third-party repositories.
Overview
MCPHub sits between MCP-compatible clients and MCP servers, giving operators one place to connect servers and expose their tools through stable endpoints. Rather than configuring each client against individual servers, teams can manage routing and access centrally.
It is suited to people running several MCP servers who want a self-hosted dashboard, shared controls, and operational visibility. It supports both local and remote connections and offers file-based configuration by default, with PostgreSQL available for database mode.
Key Features
- Routes clients to all connected servers, a group, an individual server, or smart routing endpoints.
- Connects to local and remote servers over stdio, SSE, and Streamable HTTP.
- Manages authentication, bearer keys, OAuth 2.0, and server or group visibility.
- Supports per-user credentials with encrypted storage and isolated stdio runtimes.
- Provides a web dashboard, activity logs, server logs, and health checks.
- Allows server configuration changes without downtime through hot-swappable configuration.
- Offers PostgreSQL-backed storage and vector-based smart routing as optional capabilities.
Use Cases
- A developer running several MCP servers can give Claude Code, Cursor, or another compatible client a single gateway endpoint.
- A team can group servers by project or purpose and control which users can access each group.
- An operator can centralize request logs and health checks when troubleshooting a self-hosted MCP setup.
- A deployment that needs individual users' credentials for a shared server can use per-user credential bindings.
Project Facts
- Language: TypeScript
- License: Apache-2.0
- Stars: 2.5k
- Forks: 330
- Topics: mcp, mcp-gateway, mcp-hub, mcp-router, mcp-server
- Archived: No
Getting Started
Run the Docker image with a persistent data directory:
docker run -p 3000:3000 -v ./data:/app/data samanhappy/mcphub
Open http://localhost:3000 to access the dashboard. See the README and documentation for setup, configuration, and client connection details.
Alternatives
- mcp-gateway: A Rust gateway that also exposes REST APIs and uses on-demand tool discovery, while MCPHub focuses on managing MCP server connections and access.
- magictunnel: A Rust MCP proxy centered on natural-language tool discovery and invocation, rather than MCPHub’s centralized routing, access controls, and operations.
Considerations
- Docker is the recommended quick-start option. Running from source requires Node.js 20 or newer and pnpm 10.12.4.
- MCP endpoints require authentication by default. Disabling it should be limited to trusted environments.
- The default file-based configuration suits straightforward deployments; PostgreSQL is available for database mode.
- Smart routing relies on vector semantic search and is listed with PostgreSQL and pgvector in the project stack.
- The repository is active and licensed under Apache-2.0. Review the project's configuration and security guidance before exposing a gateway outside a trusted network.
Source repository
Open the original repository on GitHub.
17 counted GitHub visits
Related repositories
Similar repositories that may be relevant next.

OrcaReplay: Record and Replay AI Agent Runs
October 2, 2026
OrcaReplay records AI agent runs so developers can inspect what happened, replay interactions offline, and fork a run from a checkpoint onto another model. It is a TypeScript CLI for debugging and comparing agent behavior without modifying the agent.

openmake_llm: Coordinate AI Models, Agents, and Tools
October 1, 2026
OpenMake is a self-hosted AI workspace that coordinates chat models, agents, and tools in one place. It suits people who want multi-step AI workflows with local or open-weight models, while keeping infrastructure and model choices under their control.

zennotes: Manage Markdown Notes with Keyboard-First Tools
October 1, 2026
ZenNotes is a local-first Markdown notes app for people who prefer plain files, keyboard navigation, and Vim motions. It runs as an Electron desktop app or a self-hosted web app, with search, diagrams, CLI, and MCP integrations.

Kun: Local-First AI Agent Workspace
September 27, 2026
Kun is a local-first workspace for using AI agents across coding, writing, design, research, and automation. Its desktop GUI and terminal TUI share one runtime, keeping tasks and approvals connected.