WhatsApp MCP for Claude: Claude Code, Desktop and claude.ai

What Claude is, and which Claude you need

Claude is Anthropic’s AI assistant. It comes as the claude.ai website and mobile apps, the Claude desktop app, and Claude Code, a coding agent that runs in your terminal and IDE. All of them can use MCP tools, but they connect to servers differently, and WaBot’s server needs two custom HTTP headers (clientid and clientsecret) on every request. That one detail decides the route:

First, get your credentials: in the WaBot app open Developer → Developer Credentials. You’ll see your Client ID, Client Secret and a Copy MCP config button. Treat the secret like a password.

Set up Claude Code (recommended)

Anthropic’s Claude Code MCP documentation adds a remote server with claude mcp add --transport http and passes headers with --header. Run this in your terminal, with your own values in place of the placeholders:

claude mcp add --transport http wabot https://api.wabot.shop/mcp \
  --header "clientid: <Client ID>" \
  --header "clientsecret: <Client Secret>"

• By default this is saved at local scope: only you, only the current project, stored in ~/.claude.json. Add --scope user to use WaBot in every project.
• Start Claude Code and type /mcp to check that wabot is connected and see its tools.

Sharing the setup with your team

Project scope writes a .mcp.json file to the project root, which goes into version control. Never put the secret in it. Claude Code expands environment variables in headers, so each person sets WABOT_CLIENT_ID and WABOT_CLIENT_SECRET in their own shell:

{
  "mcpServers": {
    "wabot": {
      "type": "http",
      "url": "https://api.wabot.shop/mcp",
      "headers": {
        "clientid": "${WABOT_CLIENT_ID}",
        "clientsecret": "${WABOT_CLIENT_SECRET}"
      }
    }
  }
}

Set up Claude Desktop (through the mcp-remote bridge)

Claude Desktop runs local MCP servers from a config file (MCP guide for Claude Desktop). The open-source mcp-remote package acts as that local server and forwards every call to WaBot with your headers. You need Node.js 18 or newer.

1. In Claude Desktop, open the Claude menu → Settings… → Developer → Edit Config. This opens claude_desktop_config.json (on Windows, %APPDATA%\Claude\claude_desktop_config.json; on macOS, ~/Library/Application Support/Claude/claude_desktop_config.json).
2. Add the wabot entry below and save.
3. Quit Claude Desktop completely and open it again. Click the + (“Add files, connectors, and more”) button in the chat box → Connectors → Manage connectors and select wabot to see its tools.

{
  "mcpServers": {
    "wabot": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://api.wabot.shop/mcp",
        "--transport", "http-only",
        "--header", "clientid:${WABOT_CLIENT_ID}",
        "--header", "clientsecret:${WABOT_CLIENT_SECRET}"
      ],
      "env": {
        "WABOT_CLIENT_ID": "<Client ID>",
        "WABOT_CLIENT_SECRET": "<Client Secret>"
      }
    }
  }
}

Why no space after the colon in clientid:${WABOT_CLIENT_ID}? mcp-remote’s README notes that Claude Desktop on Windows doesn’t escape spaces inside args, so the values live in env and the header strings have no spaces. This file stays on your computer; don’t sync it to a shared drive.

claude.ai custom connectors: where it stands

On claude.ai you add a server under Customize → Connectors → Add custom connector. Per Anthropic's connector docs, fixed credentials go in a Request headers section that is in beta for a limited set of organisations, and Anthropic reviews each custom header name before Claude will send it. WaBot uses the custom names clientid and clientsecret, not a standard Authorization header, so unless your organisation has the beta and those names are approved, the save fails. Use Claude Code or Claude Desktop until then. ChatGPT users: see the ChatGPT and Codex guide.

Use case 1: triage today’s WhatsApp chats

Good first test, because nothing gets sent.

List my WhatsApp chats from today. For each one, read the last few messages and tell me who is still waiting for a reply from us and what they asked. Don't send anything.

• Tools Claude calls: list_recent_chats, then get_messages for each conversation.
• What you see: a short list, for example “Asha (+91 98xxx xxxxx) asked at 11:40 if Spice Route Café is open on Sunday, no reply yet”. Ask Claude to draft replies; if the customer wrote within 24 hours, it can send one with send_session_message after you approve it.

Use case 2: create a utility template and get it approved

Create a utility template called order_ready_v1 in English on my main WhatsApp number. Body: "Hi {{1}}, your order {{2}} is ready for pickup at Spice Route Café. Reply here if you need anything." Show me the template before you submit it, then check its status with Meta.

• Tools Claude calls: list_channels (to find your number’s ID), create_template, then get_template_status with sync: true.
• What you see: the template for review, then PENDING. Meta’s review takes minutes to hours, so ask again later; Claude reports APPROVED, or REJECTED with Meta’s reason and a suggested fix (update_template resubmits). Unsure about categories? Read template categories.

Use case 3: draw an existing workflow before you change it

List my WhatsApp workflows. Then show me a diagram of the one called Price enquiry and explain in plain words what it does.

• Tools Claude calls: list_workflows, then preview_workflow_schematic.
• What you see: a Mermaid flowchart and a step list (trigger, match text, template sent). If your Claude app doesn’t draw Mermaid, paste the code into any Mermaid viewer. From here you can ask for a change; WaBot’s update_simple_autoflow edits simple auto-replies.

Stay safe

1. Ask Claude to show the message and the recipient before sending, and approve tool calls one at a time.
2. Test on your own number first.
3. Message customers only within WhatsApp’s rules: templates outside 24 hours, marketing only with opt-in.
4. New automations are created off; switch them on only after you’ve read the diagram.

Limits

• 60 requests a minute per Client ID; every tool call is written to WaBot’s audit log.
• wait_for_reply waits 2 to 30 seconds. Reply notifications from subscribe_replies only arrive in clients that keep MCP’s event stream open; we haven’t relied on Claude for that, so use wait_for_reply.
• Not available through MCP: campaigns and broadcast scheduling, campaign analytics, segments, deleting workflows, and advanced workflow steps (branching, webhooks, agent handoff, code). Use the WaBot app for those.

Full tool reference: WaBot MCP guide. Prefer an IDE? See Cursor.

Frequently asked questions

How do I connect Claude to WhatsApp?

Use WaBot’s MCP server. In Claude Code, run claude mcp add –transport http wabot https://api.wabot.shop/mcp with two –header flags for clientid and clientsecret. In Claude Desktop, add the mcp-remote bridge to claude_desktop_config.json.

Can I add WaBot as a custom connector on claude.ai?

Only if your organisation has the Request headers beta and Anthropic has approved the header names clientid and clientsecret. Otherwise use Claude Code or Claude Desktop.

Can Claude send WhatsApp messages by itself?

Yes, once connected it can send templates and, within 24 hours of a customer’s message, normal replies. Keep tool approvals on so you see each message before it goes.

Does Claude see all my WhatsApp chats?

It can read recent conversations through WaBot when you ask it to. Every tool call is logged in WaBot’s audit log.