teamhub-mcp-server
A remote MCP server enabling multiple Claude Code sessions to collaborate via shared presence, project-scoped notes, tasks, and memory, with owner-approved access control.
README
teamhub-mcp-server
A remote MCP (Model Context Protocol) server that lets multiple people's Claude Code sessions collaborate — see each other's live status, share and search project notes, hand off tasks, and share small durable facts — with per-project, owner-approved access control.
One person hosts it. Everyone else connects to it as a normal remote MCP server with a personal bearer token.
How it works
Person A's Claude Code --.
+--> teamhub-mcp-server (one Node process, one SQLite file)
Person B's Claude Code --'
- Presence is global — everyone can see who's online and what they're doing.
- Everything else (notes, tasks, memory) lives inside a project, which is permissioned. A project's owner has full access; everyone else needs an approved join request at one of two levels:
read_only— can view notes, tasks, and memorycollaborate— can also post notes, create/claim tasks, and write memory
Nothing is shared by default. Every cross-person data flow is an explicit tool call.
Prerequisites
- Node.js >= 20 (tested on 20 and 24)
- npm
- A host machine with persistent disk (VPS, Fly.io, Render, Railway — not classic serverless like Lambda/Vercel functions)
Quick start
1. Clone and install
git clone https://github.com/murali2212/-teamhub-mcp-server.git
cd -teamhub-mcp-server
npm install
2. Build
npm run build
This compiles TypeScript from src/ into dist/.
3. Add users
Each person who will connect needs a bearer token. Generate one per person:
node scripts/add-user.mjs alice
node scripts/add-user.mjs bob
Each command:
- Generates a random token
- Stores only its SHA-256 hash in
users.json(never the raw token) - Prints the raw token once — send it to that person privately
- Prints the exact
claude mcp addcommand they need to run
If a username already exists, the script refuses and tells you to remove the entry manually first (to rotate a token).
4. Start the server
npm start
The server listens on port 3000 by default. Override with the PORT environment variable:
PORT=8080 npm start
5. Each person connects (on their own machine)
Each person runs the command printed by add-user.mjs, which looks like:
claude mcp add --transport http teamhub https://<your-server-url>/mcp \
--scope user \
--header "Authorization: Bearer <their-token>"
--scope user makes it available across every project they open in Claude Code. They verify it works with /mcp inside any session.
Environment variables
| Variable | Default | Description |
|---|---|---|
PORT |
3000 |
HTTP listen port |
TEAMHUB_DB_PATH |
<project-root>/teamhub.db |
SQLite database file location |
TEAMHUB_USERS_PATH |
<project-root>/users.json |
Users/tokens file location |
Deployment
Requirements
- Persistent local disk — SQLite needs a real filesystem. Do NOT deploy to classic serverless (Vercel functions, Netlify functions, AWS Lambda) where the filesystem is ephemeral.
- HTTPS in production — the bearer token travels in a header on every request. Use a reverse proxy (Caddy, nginx + Let's Encrypt) or a tunnel (Cloudflare Tunnel, ngrok) for HTTPS.
- One long-running process — this is not a request-at-a-time function; it's a persistent Node.js server.
Docker
A multi-stage Dockerfile is included:
docker build -t teamhub-mcp-server .
docker run -d \
-p 3000:3000 \
-v /path/to/data:/app/data \
-e TEAMHUB_DB_PATH=/app/data/teamhub.db \
-e TEAMHUB_USERS_PATH=/app/data/users.json \
teamhub-mcp-server
Mount a volume for /app/data so the database and users file survive container restarts.
VPS (systemd)
# Build on the server
cd /opt/teamhub-mcp-server
npm install && npm run build
# Create a systemd service
sudo tee /etc/systemd/system/teamhub.service > /dev/null <<EOF
[Unit]
Description=TeamHub MCP Server
After=network.target
[Service]
Type=simple
User=teamhub
WorkingDirectory=/opt/teamhub-mcp-server
ExecStart=/usr/bin/node dist/index.js
Restart=always
Environment=PORT=3000
[Install]
WantedBy=multi-user.target
EOF
sudo systemctl enable teamhub
sudo systemctl start teamhub
Put Caddy or nginx in front for automatic HTTPS.
Fly.io
fly launch --no-deploy
fly volumes create teamhub_data --size 1
# Edit fly.toml to mount the volume and set env vars
fly deploy
Railway / Render
Point at this repo, set the build command to npm install && npm run build, start command to npm start, and add a persistent disk mounted where TEAMHUB_DB_PATH and TEAMHUB_USERS_PATH point.
All 18 tools
Presence (global, no project needed)
| Tool | What it does |
|---|---|
teamhub_set_status |
Update your live status so teammates can see what you're doing |
teamhub_list_presence |
See everyone's current status |
Projects (access control)
| Tool | What it does |
|---|---|
teamhub_create_project |
Create a new isolated project (you become the owner) |
teamhub_request_to_join |
Ask a project's owner for access (files a pending request — grants nothing) |
teamhub_list_join_requests |
List pending requests for projects you own |
teamhub_respond_to_join_request |
Approve or decline a join request (owner only) |
teamhub_revoke_access |
Revoke a member's access immediately (owner only) |
teamhub_list_my_projects |
List projects you own, are a member of, or have pending requests for |
teamhub_project_audit_log |
View the access-change audit trail (owner only) |
Notes (project-scoped)
| Tool | Access needed | What it does |
|---|---|---|
teamhub_post_note |
collaborate | Post a note to the project timeline |
teamhub_get_notes |
read_only | Read recent notes, newest first |
teamhub_search_notes |
read_only | Search notes by keyword (case-insensitive) |
Tasks (project-scoped)
| Tool | Access needed | What it does |
|---|---|---|
teamhub_create_task |
collaborate | Create a task, optionally assign to a teammate |
teamhub_list_tasks |
read_only | List tasks, optionally filtered by status/assignee |
teamhub_claim_task |
collaborate | Claim an open task for yourself |
teamhub_update_task |
collaborate | Update task status (open/claimed/done) with optional result |
Memory (project-scoped key-value)
| Tool | Access needed | What it does |
|---|---|---|
teamhub_remember |
collaborate | Store a key-value fact (upserts) |
teamhub_recall |
read_only | Look up a key, or list all keys |
Security
- Tokens hashed at rest —
users.jsonstores SHA-256 hashes, never raw tokens. A leaked file doesn't give working credentials. - Per-project access control — connecting to the server exposes nothing. Access must be explicitly requested and approved per project.
- Audit trail — every access-affecting event is logged and queryable by the project owner.
- Rate limiting — basic in-memory rate limiter (120 requests per 60-second window per token) guards against runaway loops.
- No internal error leaks — unhandled exceptions return a generic 500, never stack traces.
- Prompt injection mitigation — tool descriptions instruct the calling model to treat shared content as reference data, never as instructions.
Why not OAuth?
Static bearer tokens are a deliberate choice for the scale this targets (2-15 trusted friends/teammates). OAuth 2.1 with dynamic client registration is real infrastructure — disproportionate to "a handful of people I personally invite." If this ever needs self-service signup or untrusted users, that's the point to add OAuth.
Architecture decisions
- Stateless HTTP transport — a fresh MCP server instance per request, no session IDs. Makes the server trivially restartable and safe behind a load balancer. Trade-off: no server-initiated push notifications (pull-based only).
- SQLite (WAL mode) — zero-ops, single file, right-sized for small teams. Not meant to scale past ~15 people.
- AsyncLocalStorage for caller identity — tool handlers get the caller's username via
getCaller()without threading it through every function. - Per-request users.json reload — adding a new teammate doesn't require a server restart.
Project structure
teamhub-mcp-server/
src/
index.ts # Express app, /mcp endpoint, /health, error handling
auth.ts # Bearer token auth middleware, rate limiting
context.ts # AsyncLocalStorage for per-request caller identity
db.ts # SQLite setup, schema, logAudit helper
access.ts # requireProjectAccess helper, AccessError class
tools/
presence.ts # teamhub_set_status, teamhub_list_presence
projects.ts # Project CRUD, join requests, revoke, audit log
notes.ts # Post, get, search notes
tasks.ts # Create, list, claim, update tasks
memory.ts # Remember and recall key-value facts
scripts/
add-user.mjs # CLI to generate tokens and add users
Dockerfile # Multi-stage Docker build
package.json
tsconfig.json
License
MIT
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.
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.
E2B
Using MCP to run code via e2b.