MCP Hub
Aggregates dozens of pre-built connectors into one OpenAI-compatible endpoint for AI clients, with a web UI for installing, configuring, monitoring, and managing them.
README
MCP Hub — All-in-One MCP Server & Connector Ecosystem
Self-hosted hub that aggregates MCP (Model Context Protocol) connectors into a single endpoint for your AI clients — with an App-Store-style web UI, one-click installs, live logs and health monitoring.
Claude Desktop / Cursor ──Bearer key──▶ ┌─────────────────────┐
│ MCP Hub (:3000) │──▶ github (npx)
│ /mcp /api Web UI │──▶ postgres (npx)
Web browser ──session cookie──▶ └─────────┬───────────┘──▶ slack (npx) …
▼
PostgreSQL 16
Quick start
A. Local / plain Docker
cp .env.example .env
# edit .env — set POSTGRES_PASSWORD (and optionally MASTER_ENCRYPTION_KEY / SESSION_SECRET)
docker compose up -d --build
B. Portainer (web editor)
- Stacks → Add stack → Web editor and paste
portainer-stack.yml. - Add an environment variable
POSTGRES_PASSWORD(under Environment variables) before deploying. - Deploy — the image is pulled from Docker Hub (
xeniak10/mcp-hub:latest), no build needed.
Open http://<host>:3000 — on first visit you'll be asked to create the admin account (bootstrap). Then:
- Go to Marketplace, install e.g. GitHub, paste a token → the connector's child process starts in the background.
- Go to Settings → create an API key.
- Point your AI client at the hub:
{
"mcpServers": {
"hub": {
"url": "http://localhost:3000/mcp",
"headers": { "Authorization": "Bearer mcp_YOUR_KEY" }
}
}
}
All connector tools appear namespaced as github__create_issue, postgres__query, …
What you get
- Marketplace — 71 built-in connectors for developers (GitHub, GitLab, Postgres,
Kubernetes, Sentry…) and business teams — CRM (Salesforce, HubSpot, Pipedrive),
ERP/accounting (Odoo, QuickBooks, Xero), helpdesk (Zendesk, Freshdesk, ServiceNow),
communication (Microsoft 365, Email IMAP/SMTP, Twilio, Slack, Teams files),
productivity (Google Sheets, Notion, Airtable) and observability (Dynatrace).
One-click install, schema-driven config form, background dependency startup via
npx/uvxinside the hub container. Search, category filters and sorting (official first / most installed / name). - Community connector repos — paste a public GitHub repo URL in the Marketplace
("Community repos" section) and every valid connector manifest in it is fetched,
validated and merged into your catalog (
community:owner/repo/<id>namespaced, cached offline in the data volume). Anindex.jsonat the repo root lists files explicitly; otherwise root-level*.jsonare probed. New entries appear after a hub restart. - Custom connectors — drop a JSON manifest into
./connectors/(or theconnectorsvolume) and it appears in the marketplace after a restart. Seeconnectors/example.json.example. - Dashboard — live status badges (running / starting / error / stopped), stats tiles, instance search, bulk Restart all / Stop all with confirmation.
- Connector detail — rename instances inline, visual config editor, live tool
preview (
tools/listthrough the running process) with click-to-copy tool names, real-time log viewer (WebSocket tail) with level/text filters, follow mode and log download. - Light & dark theme — toggle in the sidebar; persisted per browser.
- Keyboard shortcuts —
/focuses search ·g ddashboard ·g mmarketplace ·g ssettings ·Esccloses dialogs/blurs search. - PWA-ready — installable manifest + SVG favicon.
- Backup & restore — one-click export of all installed connectors + configs to a JSON file; import restores them on a fresh instance (Settings page).
- Account management — change password (signs out all other devices), list active sessions with IP/user-agent, revoke any session.
- Failure alerts — set
ERROR_WEBHOOK_URLand the hub POSTs a JSON alert whenever a connector enters the error state (works with Slack, Discord, Teams, generic receivers; 5-minute per-connector cooldown). - Watchdog — crashed processes restart with exponential backoff; health ping every 30 s; boot recovery restores everything that was enabled before restart.
- Updates — Settings shows your current version and checks GitHub releases / Docker Hub tags every 6 h (or on Check now). When a new version is published you get a one-click copy of the exact upgrade command for your deployment type.
- Optional auto-update — enable the Watchtower sidecar with
docker compose --profile autoupdate up -d: it pulls newerxeniak10/mcp-hubimages hourly and recreates only the app container (it is label-scoped, so it never touches your database). - Monitoring & ops — unauthenticated
GET /healthzfor uptime probes;GET /api/metareports version + git commit (shown in Settings footer); API-key traffic on/mcpis rate-limited to 120 requests/min per key. - Security — AES-256-GCM encrypted configs at rest, argon2id passwords, cookie sessions, hashed API keys, login rate limiting, audit log.
Configuration
| Variable | Required | Description |
|---|---|---|
POSTGRES_PASSWORD |
✅ | Password for the bundled PostgreSQL |
MASTER_ENCRYPTION_KEY |
– | openssl rand -hex 32. Auto-generated to /app/data/master.key if empty |
SESSION_SECRET |
– | Empty = random per boot (users get logged out on restart) |
APP_PORT |
– | Host port (default 3000) |
LOG_RETENTION_LINES |
– | Ring buffer size per connector (default 5000) |
LOG_RETENTION_DAYS |
– | Log age limit (default 7) |
ERROR_WEBHOOK_URL |
– | URL to POST JSON alerts when a connector enters error state |
Volumes: pgdata (database), app-data (master key + state), npx-cache
(connector dependency cache — keeps restarts fast).
Updating
The hub cannot replace its own container, so updates are a one-liner on the host:
docker compose pull app && docker compose up -d --no-deps --force-recreate app
(Settings → Updates copies this for you when a new version is available. On Portainer use Stack → Pull and redeploy.) Want it fully hands-off?
docker compose --profile autoupdate up -d
That starts Watchtower next to the hub; every hour it checks Docker Hub and
recreates the app container when a newer image exists (--cleanup removes old
images). Your data lives in volumes, so updates are lossless.
Development
npm install
npm run dev # server (watch) + web (vite dev server on :5173, proxied /api + /mcp)
npm run typecheck # all workspaces
npm run build # shared → server → web
Stack: npm workspaces · Fastify 5 · @modelcontextprotocol/sdk · PostgreSQL 16 ·
Vite + React 18 + Tailwind CSS v4 + TanStack Query.
Layout
packages/
shared/ types shared by server & web (manifests, API DTOs, WS events)
server/ Fastify app: auth, registry (connector manifests), ConnectorManager,
unified /mcp proxy, log pipeline, SQL migrations
web/ React SPA served statically by the server in production
Adding a connector
Without rebuilding (recommended): copy
connectors/example.json.example to
connectors/<your-id>.json, fill in the manifest (command argv with {env.NAME}
placeholders + JSON-schema config form) and restart the hub. Invalid files are skipped
with a warning in the logs; custom ids that collide with built-ins are ignored.
From a GitHub repo: Marketplace → Community repos → paste owner/repo (or a full
github.com URL). The hub fetches an index.json from the repo root (array of file
paths) or falls back to root-level *.json files, validates each against the same
schema as volume manifests, and caches them under community:owner/repo/<id> ids —
so they can never shadow built-ins and keep working offline after the first fetch.
Manage/remove repos in the same section; changes apply on next restart.
Built-in: create packages/server/src/registry/connectors/<id>.ts exporting a
ConnectorManifest and register it in src/registry/index.ts. It appears in the
marketplace immediately after rebuild.
REST API summary
| Endpoint | Auth | Description |
|---|---|---|
GET /healthz |
– | Liveness probe (DB ping only, no details) |
POST /api/auth/bootstrap · login · logout |
–/session | First-run account, sign-in/out |
GET /api/meta |
session | {version, commit} of the running build |
POST /api/auth/change-password |
session | Change password; revokes all other sessions |
GET /api/auth/sessions · DELETE /:id |
session | List/revoke active sessions |
GET /api/marketplace |
session | Catalog entries with install counts |
GET/POST /api/community/repos · DELETE /:repo |
session | Manage community connector repos |
POST /api/connectors · PUT/DELETE /:id |
session | Install / rename / uninstall |
PUT /api/connectors/:id/config |
session | Save encrypted config |
POST /api/connectors/:id/enable·disable·restart |
session | Lifecycle |
POST /api/connectors/restart-all · stop-all |
session | Bulk operations |
GET /api/connectors/:id/tools |
session | Live tool list from the running process |
GET /api/logs/:id?after= + WS stream |
session | Log tail |
GET/POST /api/keys · DELETE /:id |
session | API key management |
GET /api/backup · POST /api/backup/restore |
session | Config export/import |
POST /mcp |
Bearer key or session | Unified MCP endpoint (120 req/min per key) |
All mutating endpoints write to the audit log.
Notes
- Connectors run as child processes inside the hub container (
npx -y pkg@versionoruvx). No docker socket is mounted. Python-based connectors needuvx— it is installed into the runtime image. - The MCP endpoint is stateless Streamable HTTP (
/mcp) — any HTTP MCP client works. - Resources are exposed under
connector://<slug>/<original-uri>.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.