← Back to directory

MCP Hub

Community

Gateway / proxy

A central coordinator for managing multiple MCP servers behind one unified endpoint

This gateway aggregates or proxies other MCP servers; it only becomes useful once you connect real servers behind it.

How to install ↓ Compare GitHub source repository ↗

Category: Dev Tools #74 of 438
Stars: ★ 521 Popular
Transport: stdio (local process) · SSE · Streamable HTTP
Runtime: Node.js 18+
Credentials: No credential needed
License: MIT
Last commit: 14 months ago ⚠

63FMRS · C

MCP Hub focuses on centralized orchestration and monitoring of multiple MCP servers, with highlights including a single /mcp endpoint aggregating all capabilities, automatic namespacing to avoid conflicts, real-time SSE events, and VS Code config compatibility. It does not implement Roots, Sampling or Completion, and it ships no domain-specific tools of its own; note the security surface created by ${cmd:} command execution and secret injection.

Strongest · Documentation 16/20 Weakest · Security and permissions 9/20

Reliability

10/20

Security and permissions

9/20

Maintenance

14/20

Documentation

16/20

Setup experience

14/20

Why each score

Reliability 10/20: Static review sees README only; no cached manifest at repository root and no CI workflows or committed tests are evidenced. The described surface (unified /mcp endpoint, namespacing, SSE events, auto-reconnect with backoff, health checks, typed error classes) is coherent, but every claim rests on self-authored docs and cannot be verified against real init handshake or tool-list behavior. No reproducible failure-mode evidence or dependency-control detail. Given the static cap of 12 and thin verifiable evidence, score 10: happy path plausible, verification and edge/error coverage unproven.
Security and permissions 9/20: No malware, credential theft, or exfiltration intent visible. Risks are by design: ${cmd: ...} placeholders execute arbitrary local commands (e.g. op/aws) and inject output into env and headers; MCP_HUB_ENV broadcasts secrets into every managed server; the /api/* REST surface exposes start/stop/restart, tool execution, and resource access with no documented bind address, localhost restriction, or auth, and no confirmation gate for command execution or tool invocation. OAuth 2.0 with PKCE is a positive. No red line triggered, but least privilege, confirmation, and data-flow disclosure are incomplete, so 9.
Maintenance 14/20: MIT license is explicit; 517 stars and 21 open issues, with a clear owner (ravitemer) and a coherent project family (mcphub.nvim, mcp-registry) plus a partially completed Todo indicating ongoing work. However there is no dated release history, no dependency-update policy, and no security-response channel in the provided material. Fits 'active but governance/versioning gaps,' score 14.
Documentation 16/20: Documentation is strong: install, CLI options, STDIO and remote server fields, VS Code compatibility and variable substitution, multiple REST endpoints with response samples, SSE event types and payloads, error classes, log paths, architecture sequence diagrams, and registry notes. Deductions: no compatibility matrix beyond a protocol version, no auth/authorization or rate-limit guidance, cost and troubleshooting handled only in passing, and much content is unverifiable self-description. Score 16.
Setup experience 14/20: Path is clear: npm i -g mcp-hub, then launch with --port and --config, plus a single-endpoint client config example; Nix flake and Neovim integration examples lower the barrier further. Deductions: users must author their own config, dev-mode cwd must be an absolute path, VS Code ${input:} requires manually setting MCP_HUB_ENV, and there is no single end-to-end from-zero tutorial or troubleshooting for common connection failures. Under the static cap of 15, score 14.

Static review · not runListed 2026-09-10

Read the FMRS scoring method →

Fit and risk

What it can accessRuns commands or codeUses the network

Best for

Developers orchestrating several MCP servers at once
Users who want a single endpoint configured in their MCP clients
Automation that needs live status and capability-change notifications

Not for

Users who only need one simple MCP server and no middle layer
Clients requiring Roots, Sampling or Completion capabilities (not supported)
Environments without Node.js 18+

Required permissions

Read and parse the specified JSON configuration files
Execute local commands through ${cmd:} placeholders while resolving configuration
Inject MCP_HUB_ENV and per-server env variables into managed child processes
Spawn stdio child processes using the configured command/args
Make network requests to configured remote URLs with custom auth headers
Write log and workspace cache files under the XDG state directory

Risks and side effects

${cmd: command args} executes arbitrary local commands during config resolution, which is risky if the config is tampered with
Starting local stdio servers is equivalent to running arbitrary commands with your user privileges
MCP_HUB_ENV and server env values may hold secrets that can leak if configs or logs are readable by others
Remote server credentials (OAuth, Bearer tokens) must be stored securely
Misconfigured --auto-shutdown can terminate the hub unexpectedly

Setup

Before you start

Runtime：Node.js 18+

MCP_HUB_ENV optionalsecret A JSON-string env var that injects shared key-value pairs (like ${input:} variables or tokens) into every managed MCP server; its contents may include secrets — set it only when needed.

Install globally: npm install -g mcp-hub.
Prepare a JSON config file (JSON5 supported, VS Code servers/mcpServers keys and ${} placeholders supported).
Start the hub: mcp-hub --port 3000 --config path/to/config.json (repeat --config to merge multiple files in order).
Point your MCP client at the single endpoint, e.g. for Claude Desktop: {"mcpServers":{"Hub":{"url":"http://localhost:37373/mcp"}}}.

Requires Node.js >= 18.0.0.

claude_desktop_config.json

{
  "mcpServers": {
    "Hub": {
      "url": "http://localhost:37373/mcp"
    }
  }
}

Shown for Claude Desktop. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json

{
  "servers": {
    "Hub": {
      "url": "http://localhost:37373/mcp"
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal

claude mcp add --transport http Hub http://localhost:37373/mcp

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

Add the endpoint http://localhost:37373/mcp to your client and check that namespaced tools (e.g. server__tool) appear in the tool list; alternatively run GET /api/health and confirm the hub state is ready.

Troubleshooting

If the hub fails to start, check that Node.js is >= 18.0.0
If clients cannot connect, verify the port and the /mcp path (example default is 37373)
If configuration is ignored, confirm the --config paths and remember later files override earlier ones
If variables or commands do not resolve, check the ${} placeholder syntax and `${cmd:}` output
For port conflicts, inspect the workspace cache at $XDG_STATE_HOME/mcp-hub/workspaces.json and clean stale entries
For diagnostics, check logs at $XDG_STATE_HOME/mcp-hub/logs/mcp-hub.log

Things to try

Once connected, you can ask your AI assistant things like:

List all MCP servers currently connected to the Hub and their available tools
Call the filesystem server's search tool through the Hub
Start the myserver server from the config and show its status
Refresh the tools, resources, and prompts of all managed servers

Use cases

Connect many MCP servers to a client such as Claude Desktop through one endpoint

Manage startup, shutdown and configuration of local (stdio) and remote (streamable-http/SSE) MCP servers

Monitor server status and capability changes in real time via REST API or SSE events

Reuse the same mcp.json across VS Code and MCP Hub

Call MCP tools and resources from Neovim via mcphub.nvim

Supported clients

Claude Desktop

Cline

Neovim

Listed from the project's documentation, not tested by this site.

Overview

MCP Hub acts as a central coordinator between MCP servers and clients, exposing two interfaces: a management interface (/api/*) for controlling multiple MCP servers through a unified REST API and web UI, and an MCP server interface (/mcp) that lets any MCP client reach all managed server capabilities through a single endpoint. It implements the MCP 2025-03-26 specification, supports stdio, streamable-http and SSE transports, OAuth 2.0 with PKCE and header-based tokens, and adds server lifecycle management, health monitoring, auto-reconnection, hot reload, SSE-based real-time events, and VS Code configuration compatibility.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 63.3k · Tools 2 Compare with this →

NPM Sentinel MCP Server 79 · B

AI-powered NPM package analysis MCP server

★ 18 · Tools 19 Compare with this →

MartinLoop MCP Server 79 · B

An independent control layer that gives AI coding agents budgets, brakes, and receipts

★ 413 Compare with this →

GitHub MCP Server 76 · B

GitHub's official MCP server for managing repos, issues, PRs, and workflows via natural language

★ 33.7k · Tools 15 Compare with this →

Compare with other MCP servers

Source revision 9c7670a4c341 Data synced 2026-10-19 Read the FMRS scoring method