Skip to main content
Glama
ABrauser

WhatsApp MCP Server (Docker Edition)

by ABrauser

WhatsApp MCP Server (Docker Edition)

A WhatsApp MCP (Model Context Protocol) server that runs as a Docker container and is accessible over HTTP/SSE. Based on whatsapp-mcp-ts with the transport layer changed from stdio to SSE for remote access.

Features

  • 8 MCP Tools: search_contacts, list_messages, list_recent_messages, list_chats, get_chat, get_message_context, send_message, search_messages

  • Docker-first: Runs as a container, accessible via IP

  • Streamable HTTP Transport: stateless MCP over HTTP POST

  • Bearer-token auth for /sse (set MCP_AUTH_TOKEN)

  • Persistent Data: SQLite DB (WAL + FTS5 search) and WhatsApp auth in Docker volumes

  • Health Check: /health returns 503 when WhatsApp is disconnected

  • Multi-Arch: Builds for both amd64 and arm64

Related MCP server: WAHA MCP

Quick Start

1. Configure secrets

cp .env.example .env
# Generate a strong token and write it to .env:
echo "MCP_AUTH_TOKEN=$(openssl rand -hex 32)" >> .env

2. Pull and Run

# Pull the image
docker pull ghcr.io/YOUR-USERNAME/whatsapp-mcp-docker:latest

# Run with docker compose
docker compose up -d

# Watch logs for QR code
docker logs -f whatsapp-mcp

3. Scan QR Code

On first start, a QR code URL will appear in the logs. Open it in your browser and scan with WhatsApp (Settings → Linked Devices → Link a Device).

4. Configure MCP Client

This server speaks Streamable HTTP MCP (POST /sse), not legacy SSE. The exact config key depends on your client:

Gemini CLI (~/.gemini/settings.json — the key is httpUrl, not url):

{
  "mcpServers": {
    "whatsapp": {
      "httpUrl": "http://192.168.0.101:3010/sse",
      "headers": {
        "Authorization": "Bearer YOUR_TOKEN_FROM_DOTENV"
      }
    }
  }
}

⚠️ Common pitfall: Using "url" in Gemini CLI selects the legacy SSE transport, which this server does not implement. You will get 401 errors that look like an auth problem but are actually a transport mismatch. Use "httpUrl".

Claude Desktop / Cline / Continue / Cursor (most use url for Streamable HTTP):

{
  "mcpServers": {
    "whatsapp": {
      "url": "http://192.168.0.101:3010/sse",
      "headers": {
        "Authorization": "Bearer YOUR_TOKEN_FROM_DOTENV"
      }
    }
  }
}

Contact Name Resolution

WhatsApp Web (Baileys) does not give us your phone-side address book directly. To still show your saved names — not the contact's self-chosen push names — the server uses three layers, in this order:

  1. Manual override (contact_overrides.json, see below) — wins everything.

  2. Auto-linking — every incoming message's pushName is captured and stored as the notify of the corresponding @s.whatsapp.net contact. This lets the existing fuzzy-match in queries automatically connect the @lid form (used in groups & status updates) to the @s.whatsapp.net row that holds your saved address-book name. After enough message activity, most contacts resolve themselves.

  3. Fallback — name → notify → phone digits → JID.

Manual overrides (contact_overrides.json)

Use this only for contacts that auto-linking can't fix (e.g. a contact who has never written you directly, or whose push name and saved name are completely different and the fuzzy match doesn't catch it).

Path: <data dir>/contact_overrides.json (typically /opt/docker/whatsapp-mcp/data/contact_overrides.json on the host)

{
  "210444891463794@lid": "Tammy",
  "11390320418995@lid": "Amelie",
  "4917697335710@s.whatsapp.net": "Tammy"
}
  • The override wins over the DB-resolved name in every MCP response (chat_name, sender_name, last_sender_name).

  • File is hot-reloaded on save — no container restart required.

  • Each entry is a JID → display name mapping. Find a contact's JID via search_contacts or list_chats.

  • See contact_overrides.example.json in the repo for the format.

Security

  • The container runs as the unprivileged node user (UID 1000), not root. Everything it writes (/app/data, /app/auth_info) must be owned by UID 1000 — the server refuses to start otherwise and prints the exact path that is not writable.

    • Fresh named volumes (default docker-compose.yml) inherit the ownership from the image automatically.

    • Bind mounts (e.g. the Portainer stack) need a one-time owner change on the host. Change the owner only — the group (and any setgid/shared-group scheme you may use under /opt/docker) stays as it is:

      sudo chown -R 1000 /opt/docker/whatsapp-mcp/data /opt/docker/whatsapp-mcp/auth_info

      If your host user already has UID 1000, this simply makes the data yours (e.g. contact_overrides.json becomes editable without sudo). If the container must run as a different UID, set user: "<uid>:<gid>" in the compose file instead — the startup check only cares about writability, not the specific UID.

    • Upgrading from an older (root-based) image with existing data: fix ownership before pulling the new image, otherwise the container crash-loops on start (your session stays intact, it just cannot be written). Bind mounts: command above. Named volumes:

      docker compose down
      docker run --rm -v whatsapp-mcp-docker_whatsapp_data:/d -v whatsapp-mcp-docker_whatsapp_auth:/a alpine chown -R 1000 /d /a
      docker compose pull && docker compose up -d

      (Volume names are <project>_whatsapp_data / <project>_whatsapp_auth; check with docker volume ls.)

  • Always set MCP_AUTH_TOKEN unless you bind the container to 127.0.0.1 only. Without a token, anyone on your network can read your chats and send messages from your account.

  • Without TLS, the token is sent in plaintext. For LAN-only deployments this is usually acceptable; for anything wider, terminate TLS in a reverse proxy (Caddy, Traefik, nginx).

  • Logs in /app/data/wa-logs.txt and /app/data/mcp-logs.txt may contain message snippets. Treat the data volume as sensitive.

  • The QR code from initial pairing is logged at info level once — it grants full session access for a few seconds. Don’t share that log.

Endpoints

Endpoint

Method

Auth

Description

/sse

POST

Bearer

Streamable HTTP endpoint for MCP (stateless)

/health

GET

none

200 when WhatsApp connected, 503 otherwise

MCP Tools

Tool

Description

search_contacts

Search contacts by name or phone number

list_messages

Get message history for a chat (paginated)

list_chats

List chats sorted by activity or name

get_chat

Get details of a specific chat

get_message_context

Get messages around a specific message

list_recent_messages

Cross-chat time-window query (e.g. "last 8 hours") with optional ISO since/until

send_message

Send a text message to a user or group

search_messages

Full-text search across messages

Environment Variables

Variable

Default

Description

MCP_PORT

3010

HTTP server port

MCP_AUTH_TOKEN

(unset)

Bearer token required on /sse. If unset, endpoint is open.

LOG_LEVEL

info

Log level (debug, info, warn, error)

WHATSAPP_MCP_DATA_DIR

/app/data

SQLite database directory

WHATSAPP_AUTH_DIR

/app/auth_info

WhatsApp auth credentials directory

TZ

Europe/Berlin

Timezone used for human-readable timestamps

WHATSAPP_DEBUG

(unset)

If true, prints @lid mapping diagnostics on startup

Development

Requires Node.js >= 24 (stable node:sqlite and on-by-default TypeScript stripping). On Node 22/23, run with --experimental-sqlite --experimental-strip-types.

# Install dependencies
npm install

# Run locally (Node 24+)
node src/main.ts

# Run locally (Node 22/23)
node --experimental-sqlite --experimental-strip-types src/main.ts

# Type-check only
npx tsc --noEmit

# Run tests
npm test

# Build Docker image locally
docker build -t whatsapp-mcp-docker .

Dependency notes

  • @whiskeysockets/baileys 6.x pulls libsignal from git, which pins a protobufjs 6.x with critical advisories. package.json overrides lifts it to the patched 7.x runtime. test/libsignalRoundtrip.test.ts performs a full Signal handshake + encrypt/decrypt through that override so a breaking protobufjs change fails in CI instead of surfacing as Bad MAC in production.

  • Run npm audit after any dependency change; the Docker image is built with npm ci from the committed lockfile, so a rebuild alone never picks up dependency fixes.

Architecture

AI Client (Gemini CLI, Claude, etc.)
    │
    │ HTTP POST (Streamable)
    ▼
┌─────────────────────┐
│  Express Server     │  Port 3010
│  ├── POST /sse      │  Stateless MCP
│  └── GET /health    │  Health check
├─────────────────────┤
│  MCP Server         │  7 Tools
├─────────────────────┤
│  Baileys            │  WhatsApp Web API (Reconnects + Backoff)
├─────────────────────┤
│  SQLite             │  Messages & Contacts (Batch Sync)
└─────────────────────┘

Credits

Based on whatsapp-mcp-ts by jlucaso1.

Related MCP Connectors

  • Drive WhatsApp from any MCP client: pair devices, send text and media, manage contacts and groups.

  • Your own WhatsApp as an MCP server: read, search and send from any MCP client.

  • WhatsMCP connects Claude and other MCP-compatible AI agents directly to WhatsApp. Send and receive text, images, documents, and voice notes; manage groups (create, add/remove members, promote admins); look up contacts and profiles; follow channels; and read call and message history — all through a standard MCP interface. For voice use cases, WhatsMCP offers SIP-based calling plans (inbound-only, or full inbound/outbound) so AI voice agents can answer and place WhatsApp calls, plus low-latency WebSocket integrations with voice agent providers like ElevenLabs. Multiple WhatsApp accounts can be paired and managed per workspace, with webhook support for real-time inbound message delivery to your own infrastructure.

  • Let Claude or ChatGPT search, read and send your WhatsApp messages over MCP. OAuth sign-in.

Related MCP Servers

  • F
    license
    A
    quality
    D
    maintenance
    An MCP server that wraps the wacli tool to enable AI clients to read, search, and send WhatsApp messages through a personal account. It provides comprehensive tools for managing chats, groups, and contacts using an existing authenticated WhatsApp session.
    27
    -
  • F
    license
    Not graded
    quality
    Not graded
    maintenance
    A self-hosted MCP server that connects AI clients to WhatsApp via the WAHA HTTP API. It enables users to manage sessions, search contacts, and send or receive messages and media directly through natural language interfaces.
    12 npm
    -
  • F
    license
    Not graded
    quality
    B
    maintenance
    WhatsApp MCP server that exposes messaging, groups, contacts, and profile management as tools and resources for AI agents, supporting Baileys and Meta Cloud API.
    23
    -
  • A
    license
    Not graded
    quality
    C
    maintenance
    MCP server that connects AI agents to WhatsApp using the multi-device API, enabling messaging, group management, and more as a regular user.
    13 npm
    MIT