Claude cannot read WhatsApp on its own. To let it summarize a client thread, find an address someone sent you, or draft a reply, you connect it to your account through an MCP server. MCP (the Model Context Protocol) is the open standard Claude uses to talk to outside tools.
This guide uses Automata, a free, open-source WhatsApp MCP server. It pairs your WhatsApp account the same way WhatsApp Web does, then lets you decide chat by chat what Claude may read or send. By the end, Claude will be able to work with the chats you approve and nothing else.
What you need
- A WhatsApp account. Use a spare number while testing, because unofficial clients carry some risk of rate limits or bans.
- A Claude account on a plan that supports custom connectors. Anthropic changes plan availability over time, so check their current details.
- Somewhere to run Automata. The quickest option is the one-click Railway template; you can also self-host with Docker.
- About ten minutes.
How the connection works
Three pieces are involved, and only the middle one is new:
- Claude is the MCP client. It sends requests to a URL.
- Automata is the MCP server. It checks the connector token on every request, enforces which chats are allowed, and exposes five WhatsApp tools.
- WhatsApp is paired to Automata as a linked device, so Claude never touches your phone or your account credentials.
Because the permission check happens in Automata and not in the prompt, Claude cannot talk its way into a chat you did not allow. Read more about how Automata enforces permissions.
Step 1: Deploy Automata
Deploy Automata somewhere Claude can reach. If you plan to use Claude.ai or Claude Desktop, the server needs a public HTTPS URL, because those apps connect from Anthropic's servers and cannot see localhost.
Claude Code runs on your machine, so a local Automata instance at http://localhost:3000 works for it. For Claude.ai and Claude Desktop, deploy to Railway or your own server.
The fastest path is the Railway template, which provisions Automata, PostgreSQL, Redis, and the Evolution API service that handles WhatsApp. To run it yourself, follow the getting started guide.
Step 2: Pair WhatsApp
Open your Automata dashboard, create a WhatsApp account connection, and continue to the QR code. On your phone, go to WhatsApp → Settings → Linked devices → Link a device and scan it.
Wait until the dashboard shows the connection as open before you move on. Automata refuses to send while the account is disconnected, since it cannot verify the recipient.
Pairing binds a real WhatsApp account. Repeated pair and unpair cycles can trigger rate limits. Do not experiment with your primary number. The full pairing notes are in the WhatsApp connection docs.
Step 3: Choose the chats
Create a connector token for this connection. This is where Automata differs from most WhatsApp MCP servers: instead of exposing every conversation, you set each chat to one of three levels.
| Level | What Claude can do |
|---|---|
| Read | List the chat, read its history, and search inside it |
| Write | Everything in read, plus send messages to it |
| No access | The chat does not exist as far as Claude can tell |
Start narrow. Give Claude read access to one or two chats, add write access only where you want it to reply, and leave everything else off. Only the tools you enable are registered, so a read-only token cannot send anything.
Step 4: Add it to Claude
Copy the connector URL from the dashboard. The endpoint looks like this:
https://your-automata.example.com/mcp/<connector-token>
Claude.ai and Claude Desktop
- Open Settings → Connectors.
- Choose Add custom connector.
- Enter a name such as
WhatsAppand paste the connector URL. - Save, then enable the connector in a new chat.
Claude Code
Run one command, replacing the URL with your own:
claude mcp add --transport http automata \
https://your-automata.example.com/mcp/<connector-token>
If you run several Automata connections, give each a distinct name such as support-whatsapp. If your client supports custom headers, you can use the plain /mcp endpoint with an Authorization: Bearer header instead, which keeps the token out of URLs. The MCP clients docs cover both formats.
Step 5: Test it
Start with the safest possible request and confirm the connection works before asking for anything real. In a new chat with the connector enabled, ask:
Check the WhatsApp connection status.
Claude calls get-connection-status and should report the account as connected. Then try prompts that match the chats you allowed:
- Summarize what Client A asked for in WhatsApp this week.
- Find the address Maria sent in the Suppliers group.
- List every open question in the Acme project chat since Monday.
- Draft a reply in the Design team group confirming the new deadline.
Ask Claude about a chat you did not allow and it will find nothing. That is the permission layer working as intended.
Keep it safe
- Treat the connector URL like a password. Anyone with the token can use the chats it allows. If it leaks, revoke that one token and create a new one; your other clients keep working.
- Prefer header authentication when your client supports it, since URLs often end up in logs.
- Grant write access sparingly. Reading is low risk; sending messages on your behalf is not.
- Keep a spare number for experiments and avoid bulk or unsolicited messaging.
Troubleshooting
| Symptom | Likely cause and fix |
|---|---|
Claude shows a 401 error | The token is missing, invalid, expired, or revoked. Create a new connector token and update the URL. |
| Claude.ai cannot connect | The URL is not public. Claude.ai connects from Anthropic's servers, so use a deployed HTTPS URL, not localhost. |
| A chat is missing from the list | It is not allowed on this token, or it has no stored messages yet. Check the chat permissions and your history sync. |
| Read tools return a missing-prerequisite response | The Evolution database URL is not configured. See the WhatsApp setup notes. |
| Sending is refused | The WhatsApp account is disconnected, or the chat does not have write access. Re-pair the account or raise the permission. |
More fixes are in the troubleshooting guide.
Next steps
- Read the WhatsApp MCP overview for what the five tools do and how scoped access compares to other servers.
- Connect more clients, including ChatGPT and Cursor, with the MCP clients guide.
- Lock things down further with the permissions docs.