WhatsApp MCP Server

WhatsApp MCP Server

Manage WhatsApp Business templates and send messages from Claude, ChatGPT, Cursor, VS Code Copilot, or any MCP-compatible client — powered by the Meta Cloud API.

[MCP Compatible] [Meta API v24.0] [MIT License] [Python 3.10+]

What It Does

8 tools covering the full template lifecycle: create → validate → approve → send.

Quick Start

1. Clone & Install

git clone https://github.com/nakulben/whatsapp-mcp.gitcd whatsapp-mcppython -m venv venvsource venv/bin/activate   # Windows: venv\Scripts\activatepip install -r requirements.txt

2. Configure Credentials

cp .env.example .env

META_ACCESS_TOKEN=your_access_tokenMETA_WABA_ID=your_whatsapp_business_account_idMETA_PHONE_NUMBER_ID=your_phone_number_idMETA_APP_ID=your_app_id              # Optional, for media uploadsMETA_API_VERSION=v24.0               # Optional, defaults to v24.0

Environment variables are used by all modes — local stdio and hosted remote.

How to get these? Go to Meta for Developers, create or select your app, navigate to WhatsApp > API Setup.

3. Connect to Your MCP Client

The server supports 3 transport modes:

For HTTP transports, you can customize host/port:

python -m whatsapp_mcp --transport streamable-http --host 0.0.0.0 --port 8000

Claude Desktop (stdio — local)

Add to claude_desktop_config.json:

{  "mcpServers": {    "whatsapp": {      "command": "/path/to/whatsapp-mcp/venv/bin/python",      "args": ["-m", "whatsapp_mcp"],      "env": {        "META_ACCESS_TOKEN": "your_access_token",        "META_WABA_ID": "your_waba_id",        "META_PHONE_NUMBER_ID": "your_phone_number_id",        "META_APP_ID": "your_app_id"      }    }  }}

Claude.ai Web (remote — streamable-http)

Claude.ai connects to remote MCP servers as custom connectors. The connection originates from Anthropic's cloud servers, not from your machine.

1. Host the server with env vars configured, behind HTTPS:
   bash

   python -m whatsapp_mcp --transport streamable-http --host 0.0.0.0 --port 8001

2. Put it behind HTTPS using nginx, Caddy, or a tunnel (ngrok, Cloudflare Tunnel)
3. In Claude.ai: go to Customize > Connectors → Add custom connector
4. Enter your server URL (e.g. https://your-domain.com/mcp/)
5. Claude supports authless or OAuth-based servers. For simplest setup, leave auth blank — the server will use the env vars you configured in step 1.

Note: Claude.ai does not support custom request headers. The server must be pre-configured with Meta credentials via environment variables. Each hosted server serves one WhatsApp Business Account.

location /mcp/ {    proxy_pass http://127.0.0.1:8001/;    proxy_http_version 1.1;    proxy_set_header Upgrade $http_upgrade;    proxy_set_header Connection "upgrade";    proxy_set_header Host $host;    proxy_set_header X-Real-IP $remote_addr;    proxy_read_timeout 86400;}

ChatGPT (remote — Responses API)

ChatGPT supports remote MCP servers via the Responses API. It supports both Streamable HTTP and SSE transports.

Option 1 — Server pre-configured with env vars (simplest):

from openai import OpenAIclient = OpenAI()resp = client.responses.create(    model="gpt-4.1",    tools=[{        "type": "mcp",        "server_label": "whatsapp",        "server_url": "https://your-domain.com/mcp/",        "require_approval": "never",    }],    input="List all my approved templates",)

Option 2 — Per-request credentials via Bearer token:

Encode your Meta credentials as base64 JSON and pass them in the authorization field. OpenAI forwards this value as the Authorization header to your MCP server:

# Create the tokenecho -n '{"access_token":"EAA...","phone_number_id":"123","waba_id":"456"}' | base64# Output: eyJhY2Nlc3NfdG9rZW4iOiJFQUEuLi4iLCJwaG9uZV9udW1iZXJfaWQiOiIxMjMiLCJ3YWJhX2lkIjoiNDU2In0=

resp = client.responses.create(    model="gpt-4.1",    tools=[{        "type": "mcp",        "server_label": "whatsapp",        "server_url": "https://your-domain.com/mcp/",        "authorization": "eyJhY2Nlc3NfdG9rZW4iOiJFQUEuLi4iLCJwaG9uZV9udW1iZXJfaWQiOiIxMjMiLCJ3YWJhX2lkIjoiNDU2In0=",        "require_approval": "never",    }],    input="List all my approved templates",)

Note: ChatGPT only supports remote MCP servers (no local stdio). Your server must be publicly accessible over HTTPS.

Cursor (stdio)

Add to .cursor/mcp.json in your project:

{  "mcpServers": {    "whatsapp": {      "command": "/path/to/whatsapp-mcp/venv/bin/python",      "args": ["-m", "whatsapp_mcp"]    }  }}

VS Code Copilot (stdio)

Add to .vscode/mcp.json:

{  "servers": {    "whatsapp": {      "type": "stdio",      "command": "/path/to/whatsapp-mcp/venv/bin/python",      "args": ["-m", "whatsapp_mcp"]    }  }}

Per-Request Credentials (direct HTTP / curl / scripts)

For programmatic access or custom MCP clients, you can pass per-request credentials instead of relying on server env vars. Two methods are supported:

Method 1 — Bearer token (recommended):

Base64-encode a JSON object with your Meta credentials:

# Create the tokenTOKEN=$(echo -n '{"access_token":"EAA...","phone_number_id":"123","waba_id":"456"}' | base64)
# Use itcurl -H "Authorization: Bearer $TOKEN" https://your-server.com/mcp/ ...

Required fields: access_token, phone_number_id, waba_id. Optional: app_id, api_version.

Method 2 — X-Meta- headers:*

If neither Bearer token nor X-Meta-* headers are present, the server falls back to environment variables.

Usage Examples

Once connected, just talk to your AI assistant:

"Create a marketing template called summer_sale with a header image, body text about 50% off, and a Shop Now button"

"List all my approved templates"

"Send the order_confirmation template to +919876543210 with order number ORD-456"

"Validate this template before I submit it: ..."

"Check the status of template ID 123456789"

Supported Template Types

Meta's API has 2 template categories(excluding Authentication). Within each category, templates can have different structural variants — each with its own component layout and validation rules.

Marketing Templates

Utility Templates

How routing works: When you call create_template, the server inspects the components to auto-detect the structural variant (e.g., presence of cards[] → Carousel, CATALOG button → Catalog) and applies the correct validator. You just pass category: "MARKETING" or "UTILITY" — the variant is determined from the component structure.

Running Tests

pip install pytest pytest-asynciopython -m pytest tests/ -v

Project Structure

whatsapp-mcp/├── whatsapp_mcp/│   ├── __init__.py          # Package version│   ├── __main__.py          # Entry point (python -m whatsapp_mcp)│   ├── config.py            # Environment config loader│   ├── meta_api.py          # Async Meta Graph API client│   ├── middleware.py         # ASGI middleware for per-request credentials│   ├── server.py            # MCP server with 8 tools│   ├── models/              # Pydantic data models│   │   ├── body.py          # Body component│   │   ├── header.py        # Header component (text/image/video/document)│   │   ├── footer.py        # Footer component│   │   ├── buttons.py       # Button types (URL, phone, quick reply, etc.)│   │   ├── buttons_component.py│   │   ├── enums.py         # Template categories, types, formats│   │   └── order_models.py  # Order-related models (checkout templates)│   └── validators/│       ├── create/          # 12 template creation validators│       └── send/            # 11 template send validators├── tests/│   ├── test_validators.py   # Validator tests│   ├── test_meta_api.py     # API client tests (mocked HTTP)│   └── test_tools.py        # MCP tool registration & helper tests├── .env.example├── requirements.txt├── LICENSE                  # MIT└── ROADMAP.md

Requirements

• Python 3.10+
• Meta WhatsApp Business Account
• System User access token with whatsapp_business_messaging and whatsapp_business_management permissions

Dependencies

Roadmap

See ROADMAP.md for planned features.

License

MIT — see LICENSE.

Built by Jina Connect — the WhatsApp Business CX platform.