whatsapp-mcp-server

whatsapp-mcp-server

A least-privilege, send-only MCP server for WhatsApp, backed by Meta's official WhatsApp Cloud API. It exposes tools to send text messages and pre-approved templates.

Category
Visit Server

README

whatsapp-send-mcp

A least-privilege, send-only Model Context Protocol server for WhatsApp.

It exposes two send tools and nothing else, backed by Meta's official WhatsApp Cloud API:

  • send_whatsapp_message — send a free-form text message (deliverable only inside WhatsApp's 24-hour customer service window).
  • send_whatsapp_template — send a pre-approved template message (deliverable at any time, the reliable path for scheduled notifications).

That is the entire surface area:

  • It can send a text or template message to a specified or default recipient.
  • It cannot read messages, list chats, access contacts, read statuses, or download/upload media. None of that is implemented, and the Graph API token scope it uses (whatsapp_business_messaging) does not grant a "read messages" capability in the first place. Send-only is enforced by the platform, not just by this wrapper.

Why the official Cloud API (and not Baileys / whatsapp-web.js)

  • Official and Terms-of-Service compliant.
  • Does not require linking your personal WhatsApp account as a device.
  • The API has no endpoint to read a user's message history, so a send-only integration is enforced at the platform level.

What it does not do (by design)

  • No inbound message handling or webhooks.
  • No reading of message history, chats, contacts, or statuses.
  • No media upload or download.
  • No group messaging.
  • No template management from the server (creating/editing/deleting templates). Templates are created once in WhatsApp Manager, or via the included one-off create:template script; the server only sends already-approved templates.

These are deliberate non-goals. Adding a read capability later would be a separate, deliberate decision, not something bundled into this package.

Requirements

  • Node.js 20 or newer.
  • A WhatsApp Cloud API app on Meta for Developers, with an access token and a phone number ID.

Install

Run directly with npx (no install needed):

npx @praise25/whatsapp-mcp-server

Or install globally:

npm install -g @praise25/whatsapp-mcp-server

Configuration

All configuration is via environment variables. Nothing is hardcoded, and the access token is never logged.

Variable Required Description
WHATSAPP_ACCESS_TOKEN Yes Meta system-user or temporary access token, scoped to whatsapp_business_messaging.
WHATSAPP_PHONE_NUMBER_ID Yes The Cloud API phone number ID (a numeric ID, not the phone number itself).
WHATSAPP_DEFAULT_RECIPIENT No Default recipient in E.164 format (e.g. +2348012345678), used when to is omitted.
WHATSAPP_GRAPH_API_VERSION No Graph API version segment. Defaults to v21.0.

The server validates required variables at startup and fails fast with a clear error if any are missing. See .env.example.

How to get a WhatsApp Cloud API test number and access token

  1. Go to Meta for Developers and create (or open) an app of type Business.
  2. Add the WhatsApp product to the app.
  3. Open WhatsApp > API Setup. Meta provides a free test phone number for development.
  4. Copy the Phone number ID shown there into WHATSAPP_PHONE_NUMBER_ID.
  5. Copy the temporary access token into WHATSAPP_ACCESS_TOKEN. (Temporary tokens last ~24 hours. For a longer-lived token, create a System User in Meta Business Settings and grant it the whatsapp_business_messaging permission only.)
  6. Still on the API Setup page, add your own phone number to the allowed recipient list so the test number is permitted to message you.
  7. Messages sent from a test number can only go to numbers on that allowed list until the number is fully verified for production.

The one tool: send_whatsapp_message

Input:

Field Type Required Description
to string Optional if WHATSAPP_DEFAULT_RECIPIENT is set Recipient in E.164 format, e.g. +2348012345678.
message string Yes The text body to send.

Output:

  • On success: the WhatsApp message ID.
  • On failure: a distinguishable, human-readable error (authentication, invalid recipient, rate limit, server error, network error), not a generic "failed".

Behavior notes:

  • Calls POST https://graph.facebook.com/v21.0/{PHONE_NUMBER_ID}/messages with a type: "text" payload.
  • Retries at most once, and only on a 5xx or network error. It never retries on a 4xx (those are permanent: bad token, bad number, etc.).

The 24-hour window. A free-form text message is only delivered if the recipient has messaged your number within the last 24 hours. Outside that window the API still returns success, but WhatsApp drops delivery. For reliable, unprompted delivery, use a template (below).

The other tool: send_whatsapp_template

Sends a pre-approved template, which WhatsApp delivers at any time, with no 24-hour-window dependency. This is the reliable choice for scheduled notifications, alerts, and digests.

Input:

Field Type Required Description
to string Optional if WHATSAPP_DEFAULT_RECIPIENT is set Recipient in E.164 format.
templateName string Yes An approved template name (the built-in hello_world always exists).
languageCode string No BCP-47 code the template was approved in. Defaults to en_US.
headerParams string[] No Ordered values for the header's {{1}}, {{2}}, … placeholders.
bodyParams string[] No Ordered values for the body's {{1}}, {{2}}, … placeholders. Each value must be a single line (no newlines/tabs, WhatsApp rejects them inside a variable).

The server stores none of the template wording. Meta owns the approved template; the caller supplies only the variable values that fill its blanks.

Registering a template

Templates are created once in WhatsApp Manager (Create template), or reproducibly via the included one-off script:

npm run create:template

This posts a template to your WABA and requires the token to also have the whatsapp_business_management permission (sending only needs messaging). After submission, wait for the status to become APPROVED before sending it. Note WhatsApp's variable-to-text ratio rule: a body with too many variables relative to its fixed wording is rejected, add more fixed text or fewer variables.

Use with an MCP client

Claude Desktop / Claude Code

Add to your MCP client configuration (e.g. claude_desktop_config.json):

{
  "mcpServers": {
    "whatsapp-send": {
      "command": "npx",
      "args": ["-y", "@praise25/whatsapp-mcp-server"],
      "env": {
        "WHATSAPP_ACCESS_TOKEN": "your-access-token",
        "WHATSAPP_PHONE_NUMBER_ID": "1234567890",
        "WHATSAPP_DEFAULT_RECIPIENT": "+2348012345678"
      }
    }
  }
}

The server communicates over stdio.

Remote deployment (HTTP transport)

For unattended / scheduled use, the server can also run as an always-on HTTPS service and be added to Claude as a custom connector (Settings → Connectors → Add custom connector → give it the public URL). It exposes a Streamable HTTP endpoint at /mcp.

Security: the HTTP endpoint requires the secret token. A public send endpoint must be authenticated, otherwise anyone who finds the URL could send WhatsApp messages on your account. The server refuses to start over HTTP without MCP_AUTH_TOKEN set. It accepts the token two ways:

  • Header (preferred): POST /mcp with Authorization: Bearer <MCP_AUTH_TOKEN>. Use this whenever your client can set headers (e.g. Claude Code's --header). The token never appears in the URL.
  • Path (fallback): POST /mcp/<MCP_AUTH_TOKEN>. For connector UIs that only accept a URL and cannot attach a header. The URL then is the secret, treat it accordingly (see SECURITY.md): use the long hex token, do not share or screenshot the URL, and be aware the host's access logs may record the path.

Extra env vars for HTTP mode:

Variable Required Description
MCP_TRANSPORT Yes (http) Set to http to run the HTTP server (default stdio).
MCP_AUTH_TOKEN Yes Long random secret. Callers must present it as a bearer token. Generate: node -e "console.log(require('crypto').randomBytes(32).toString('hex'))".
PORT No Port to listen on. Hosts like Render set this automatically.

Endpoints:

  • POST /mcp — the MCP endpoint (bearer auth required).
  • GET /health — unauthenticated health check, also useful for a keepalive ping to stop a free host from sleeping.

Keepalive (no external pinger needed). On a free host that sleeps when idle, the server can keep itself awake by self-pinging /health on a timer. On Render this turns on automatically (it reads RENDER_EXTERNAL_URL). Elsewhere, set KEEPALIVE_URL to your public base URL. Tune the interval with KEEPALIVE_MINUTES (default 10; the host sleeps after ~15 min idle). Note a single always-on free Render service uses roughly 730 of the 750 free monthly hours.

Deploy to Render

A render.yaml blueprint is included. Push this repo to GitHub, then in Render: New → Blueprint and point it at the repo. Set the secrets (MCP_AUTH_TOKEN, WHATSAPP_ACCESS_TOKEN, WHATSAPP_PHONE_NUMBER_ID, WHATSAPP_DEFAULT_RECIPIENT) in the Render dashboard, never in the file. Render health-checks /health; a separate keepalive ping to /health keeps a free instance awake.

Local development

npm install
npm run build      # compile TypeScript to dist/
npm test           # run unit tests (fetch is mocked; no real API calls)

Manual send tests

To verify against the real Graph API before publishing, put your credentials in a local .env (copy from .env.example) and run any of these. They actually deliver messages, so use a number you control.

npm run test:send                      # free-form text to WHATSAPP_DEFAULT_RECIPIENT
npm run test:send -- +2348012345678    # text to an explicit recipient

npm run test:template                  # built-in hello_world template (works anytime)

npm run create:template                # one-off: register the example topic_digest_report template
npm run test:digest                    # send that template once approved
npm run test:digest -- "AI News" "Morning" "OpenAI shipped X • DeepMind announced Y"

test:template and test:digest demonstrate template delivery, which works regardless of the 24-hour window.

Security

See SECURITY.md. In short: the access token is read only from an environment variable, is never committed and never logged, and revoking it in the Meta dashboard immediately disables this integration entirely.

License

MIT

Recommended Servers

playwright-mcp

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.

Official
Featured
TypeScript
Magic Component Platform (MCP)

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.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

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.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
Kagi MCP Server

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.

Official
Featured
Python
graphlit-mcp-server

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.

Official
Featured
TypeScript
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

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.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured
E2B

E2B

Using MCP to run code via e2b.

Official
Featured