Kaption WhatsApp (Cloud)

What Kaption WhatsApp (Cloud) MCP server does

Kaption WhatsApp (Cloud) MCP server connects MCP-compatible AI clients to WhatsApp conversations through the Kaption browser extension. The cloud component acts as a relay: it accepts authenticated MCP requests, associates them with an account, and passes them to the extension running in a WhatsApp Web tab.

The relay does not read WhatsApp messages and does not execute the tool implementations. The extension performs the work in the local browser context and returns the result through the relay. The public service is available at mcp.kaptionai.com; mcp-ext.kaptionai.com remains available as a backward-compatible alias.

How it works

A client discovers the OAuth authorization server, registers dynamically, and starts the authorization flow. The user authenticates with a WhatsApp OTP, after which the client receives an access token. MCP requests can then use either the SSE endpoint at /sse or the Streamable HTTP endpoint at /mcp.

On the server side, a relay Durable Object receives each request and routes it to a room associated with the authenticated account. That room maintains the WebSocket connection to the browser extension, matches responses to requests, and sends results back to the MCP client. The extension must therefore be connected to WhatsApp Web for forwarded actions to run.

Setup and configuration

Use the setup guide at kaptionai.com/mcp to connect WhatsApp to a supported AI client such as Claude, ChatGPT, or Cursor. The connection uses the hosted relay, so users do not need to deploy the repository to use the service. OAuth registration and token exchange are handled through the relay endpoints.

The repository is also a Cloudflare Worker project with Next.js application routes, Durable Objects, and tests. Local development includes npm install, npm test, and npm run dev, although the development command runs the Next.js portion without the Worker routing. Production deployments are gated through GitHub Actions rather than a local wrangler deploy command.

Tools and capabilities

Kaption WhatsApp (Cloud) MCP server exposes 16 public tools through the extension:

• query searches conversations, contacts, messages, transcriptions, labels, communities, and sessions.
• summarize_conversation retrieves or generates a conversation summary.
• manage_labels handles WhatsApp Business labels.
• manage_notes reads and updates contact notes for Business accounts.
• download_media downloads media from messages.
• manage_chat archives, pins, mutes, marks chats read or unread, and manages drafts.
• manage_reminders manages personal reminders.
• manage_scheduled_messages schedules messages from Kaption's number or, in local mode, from the user's number on that computer, including group messages.
• manage_lists manages personal chat lists.
• Contact tools list contacts, retrieve contact details, and return group memberships.
• Group tools list groups and retrieve group metadata.
• export_contacts exports contacts as CSV or JSON.
• get_analytics analyzes WhatsApp activity, rankings, response times, and exports.

The local-only get_api_info tool is excluded from the cloud surface because it returns private REST connection credentials.

Limitations and notes

Kaption WhatsApp (Cloud) MCP server depends on the browser extension and an active WhatsApp Web context. The relay forwards calls but does not provide an independent WhatsApp API implementation. Its scheduled-message tool supports both Kaption's number and a local mode using the user's number on the connected computer, with the documented distinction between those modes.

The source is public under BUSL-1.1. Production deployments are recorded in a tamper-evident transparency chain, with deployment manifests signed through Sigstore and logged to Rekor. Public endpoints expose the latest deployment, chain history, and chain-integrity verification results. These deployment controls describe how the relay is shipped; they do not change the requirement for the extension to execute WhatsApp operations locally.