AI Agents
metamcp
MCP aggregator, orchestrator, middleware and gateway in one Docker — compose many servers into one endpoint
★ 2.7k starsMIT
View guide →Self-hosted MCP gateway and control plane — connect, control and operate all your MCP servers from one place
mcphub is an open-source ai agents project by samanhappy (samanhappy/mcphub on GitHub). It has 2,503 GitHub stars and 331 forks, is released under the Apache-2.0 license, and was last updated on GitHub on 8 October 2026.
List the MCP servers your team uses and who should see which servers, groups and tools.
Run the Docker image with a ./data volume for persistence (or clone for local development with pnpm).
Optionally pre-create data/mcp_settings.json with your servers, set ADMIN_PASSWORD, then log in to the dashboard as admin and finish setup.
Point clients at http://localhost:3000/mcp (or /mcp/{group}, /mcp/{server}, /mcp/$smart) and verify tool calls flow through the hub.
Stable MCP endpoints for all servers, a group, a single server, or smart routing scoped to a group.
AI-powered tool discovery using vector semantic search, so clients find the right tool across many servers.
Bind personal keys to one shared server with encrypted storage and isolated stdio runtimes per user.
OAuth 2.0 (client and server modes), bearer tokens, GitHub/Google social login (database mode), and per-server/group visibility of tools, prompts and resources.
Tool-call logs with status and latency, server log inspection, health checks, hot-swappable config and tool-result compression.
PostgreSQL database mode, a full Docker image variant with Rust/Playwright for heavier servers, and a headless mode without the dashboard.
Clone MCPHub for local development; the fastest run path is Docker in step 2.
git clone https://github.com/samanhappy/mcphub.git
cd mcphubStart MCPHub on port 3000 with persistent data, as the quick start shows.
docker run -p 3000:3000 -v ./data:/app/data samanhappy/mcphubLog in at http://localhost:3000 as admin (set ADMIN_PASSWORD, or read the generated password from the server logs on first launch), add servers, and connect clients to /mcp.
# open http://localhost:3000samanhappy maintains it; site and demo are at mcphub.app, docs at docs.mcphub.app.
Yes, Apache-2.0 licensed and self-hosted. The README offers optional paid production support from the maintainer.
No — you run it yourself. The MCP servers behind it keep their own credentials, which MCPHub stores encrypted per user.
Both aggregate MCP servers behind endpoints. MetaMCP centers on namespaces/middleware; MCPHub centers on the control plane — per-user credentials, smart routing, logs and health monitoring.
Run the docker run command, open http://localhost:3000, log in as admin, add a server in the dashboard, and connect your client to /mcp.
MCP aggregator, orchestrator, middleware and gateway in one Docker — compose many servers into one endpoint
Unla MCP Gateway: turn existing APIs and MCP servers into MCP endpoints with zero code changes.
All your agents and workspaces in one place — on every device you own.
Ultra-lightweight, open-source, self-hosted personal AI agent framework in Python with WebUI, tools, memory, MCP, multi-agent workflows, automation, and chat…
Connect your Zotero research library to Claude and other AI assistants through MCP.
FIRERPA / LAMDA — full-stack Android device control: remote desktop, automation, MITM, Frida and MCP.
The most-starred other AI Agents projects in the LibreAI catalogue — the realistic shortlist if mcphub is not the right fit.
A personal AI that really does things — on any OS, with your data staying yours.
An agentic skills framework & software development methodology that works.
Agent harness optimization system with skills, memory and security for Claude Code, Codex and more
The agent that grows with you — a Python framework for adaptive AI agents.
DeepSeek's plugin-based agent harness where everything is a plugin.
Free and open — grab the source, star the repo, and start creating.