Kaption WhatsApp

@kaptionai/mcp-extension

MCP server that lets AI assistants read and manage your WhatsApp conversations through the KaptionAI Chrome extension. Also supports WebMCP for zero-config browser-native AI tool discovery.

Claude / Cursor ──stdio──> mcp-whatsapp ──ws://localhost:7865──> KaptionAI Extension ──> WhatsApp Web
Browser AI Agent ──navigator.modelContext──> KaptionAI Extension ──> WhatsApp Web (WebMCP)

Agent skill

Teach Claude Code, Codex, Cursor and other agents how to use these tools well (search first, never send without confirming, reminders vs. scheduled messages, error handling):

npx skills add Kaption-AI/mcp-extension

The skill lives in skills/kaption-whatsapp/SKILL.md.

Setup

1. Install the extension

Install the KaptionAI Chrome Extension and enable the MCP bridge in settings.

2. Configure your AI tool

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "whatsapp": {
      "command": "npx",
      "args": ["-y", "@kaptionai/mcp-extension@latest"]
    }
  }
}

Claude Code

claude mcp add whatsapp -- npx -y @kaptionai/mcp-extension@latest

Cursor

Add to .cursor/mcp.json in your project or go to Settings > MCP Servers:

{
  "mcpServers": {
    "whatsapp": {
      "command": "npx",
      "args": ["-y", "@kaptionai/mcp-extension@latest"]
    }
  }
}

Windsurf

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "whatsapp": {
      "command": "npx",
      "args": ["-y", "@kaptionai/mcp-extension@latest"]
    }
  }
}

VS Code (Copilot)

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

{
  "servers": {
    "whatsapp": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@kaptionai/mcp-extension@latest"]
    }
  }
}

Or add to your VS Code settings.json:

{
  "mcp": {
    "servers": {
      "whatsapp": {
        "type": "stdio",
        "command": "npx",
        "args": ["-y", "@kaptionai/mcp-extension@latest"]
      }
    }
  }
}

Zed

Add to Zed settings (~/.config/zed/settings.json):

{
  "context_servers": {
    "whatsapp": {
      "command": {
        "path": "npx",
        "args": ["-y", "@kaptionai/mcp-extension@latest"]
      }
    }
  }
}

OpenAI Agents SDK (Python)

from agents import Agent
from agents.mcp import MCPServerStdio

whatsapp = MCPServerStdio(
    name="whatsapp",
    command="npx",
    args=["-y", "@kaptionai/mcp-extension@latest"],
)

agent = Agent(
    name="assistant",
    instructions="You can access WhatsApp conversations.",
    mcp_servers=[whatsapp],
)

Any MCP-compatible client

This package runs as a standard MCP server over stdio. To connect from any client:

npx -y @kaptionai/mcp-extension@latest

The server communicates via stdin/stdout using the MCP protocol. Point your client's MCP configuration to this command.

3. Open WhatsApp Web

Open web.whatsapp.com in Chrome or Edge. The extension will auto-connect to the MCP server.

Tools

query

Query WhatsApp data — conversations, contacts, messages, transcriptions, labels, and communities.

# List conversations
query {}

# Search everything
query { query: "meeting" }

# Unread only
query { unread: true }

# Look up a conversation with messages
query { id: "[email protected]" }

# Search contacts
query { query: "Alice", entity: "contacts" }

# List labels (Business accounts)
query { entity: "labels" }

# Filter by label
query { label: "Important" }

# List communities
query { entity: "communities" }

# Filter by community
query { community: "My Community" }

# Get session info
query { entity: "session" }

Parameter	Type	Description
query	string	Search text (names, messages, transcriptions)
id	string	Look up a specific conversation, contact, or label
entity	string	conversations, contacts, messages, transcriptions, labels, communities, session
limit	number	Max results (default 25, max 5000)
unread	boolean	Only unread conversations
label	string	Filter by label name or ID
community	string	Filter by community name or ID
before	string	Messages before this ISO 8601 timestamp (pagination)
after	string	Messages after this ISO 8601 timestamp (incremental sync)

summarize_conversation

Generate a summary of a conversation.

Parameter	Type	Description
conversation_id	string	The conversation ID

manage_labels

Manage WhatsApp Business labels — add, remove, create, or delete labels.

Parameter	Type	Description
action	string	add, remove, create, delete
label_name	string	Label name
label_id	string	Label ID (alternative to name)
conversation_id	string	Required for add/remove

manage_lists

Manage personal chat lists (custom lists). The personal account equivalent of Business labels — organize chats into custom categories.

# List all custom lists
manage_lists { action: "list" }

# Get a list with its chats
manage_lists { action: "get", name: "Family" }

# Create a new list
manage_lists { action: "create", name: "Work", conversation_id: "[email protected]" }

# Add a chat to a list
manage_lists { action: "add_chat", name: "Family", conversation_id: "[email protected]" }

# Remove a chat from a list
manage_lists { action: "remove_chat", name: "Family", conversation_id: "[email protected]" }

# Delete a list
manage_lists { action: "delete", name: "Old List" }

Parameter	Type	Description
action	string	list, get, create, edit, delete, add_chat, remove_chat
id	string	List ID
name	string	List name (for create/edit, or to resolve by name)
conversation_id	string	Chat ID(s) to add/remove

manage_chat

Manage chat state — archive, unarchive, mark as read/unread, pin, unpin, mute, unmute, set/clear draft messages.

# Archive a chat
manage_chat { action: "archive", conversation_id: "[email protected]" }

# Mark as read
manage_chat { action: "mark_read", conversation_id: "[email protected]" }

# Pin a chat (max 3)
manage_chat { action: "pin", conversation_id: "[email protected]" }

# Mute for 1 week
manage_chat { action: "mute", conversation_id: "[email protected]", mute_duration: "1w" }

# Set a draft message
manage_chat { action: "set_draft", conversation_id: "[email protected]", text: "Hey, I'll call you back" }

Parameter	Type	Description
action	string	archive, unarchive, mark_read, mark_unread, pin, unpin, mute, unmute, set_draft, clear_draft
conversation_id	string	The conversation ID
mute_duration	string	8h, 1w, or forever (default). Only for mute
text	string	Draft text. Required for set_draft

manage_reminders

Create and manage personal reminders. Stored in the cloud and delivered via the Kaption extension.

# List active reminders
manage_reminders { action: "list" }

# Create a reminder
manage_reminders { action: "create", title: "Follow up with client", datetime: "2026-03-07T14:00:00Z" }

# Complete a reminder
manage_reminders { action: "complete", id: "rem_abc123" }

# List all including completed
manage_reminders { action: "list", filter: "all" }

Parameter	Type	Description
action	string	list, get, create, update, delete, complete, uncomplete
filter	string	For list: active (default), completed, all
id	string	Reminder ID (for get/update/delete/complete/uncomplete)
title	string	Reminder text (max 800 chars, no newlines)
datetime	string	ISO 8601 datetime
notification_type	string	extension, whatsapp, or automatic (default)

manage_scheduled_messages

Schedule messages to be sent automatically at a specific time, in one of two modes:

• bot (default) — sent from Kaption's WhatsApp number, not yours. Stored in Kaption's cloud and sent even when your computer is off. One-to-one chats only; one line, up to 800 characters.
• local ("From this computer") — sent from your own WhatsApp number, as you, while this computer and WhatsApp are open. Nothing leaves your device. Kaption keeps it within safe limits automatically (a few messages an hour and a day, minutes apart, only to chats where the other person has written, at most 3 a day to groups) and refuses what doesn't fit, with the reason. The only mode that can send to groups (ones you can post in and posted in within the last 30 days). Text only. A message whose time passes while the computer is off is marked missed, never sent late on its own.

The local mode has to be turned on by you in Kaption first (the "From this computer" option in the scheduling picker, after reading the limits and risks). An AI assistant can't turn it on; until you do, local requests are refused with (reason: no-consent). It is also rolled out gradually: where it isn't available yet, local requests are refused with (reason: flag-off) and the bot works as before.

# List pending messages of both modes (each one has a "mode")
manage_scheduled_messages { action: "list" }

# Schedule with the Kaption bot
manage_scheduled_messages { action: "create", message: "Hey, just following up!", datetime: "2026-03-07T09:00:00Z", conversation_id: "[email protected]" }

# Schedule from your own number, or to a group
manage_scheduled_messages { action: "create", mode: "local", message: "Running 10 min late", datetime: "2026-03-07T09:00:00-03:00", conversation_id: "[email protected]" }
manage_scheduled_messages { action: "create", mode: "local", message: "Standup moved to 10", datetime: "2026-03-07T09:00:00Z", conversation_id: "[email protected]" }

# Cancel a scheduled message
manage_scheduled_messages { action: "delete", id: "msg_abc123" }
manage_scheduled_messages { action: "cancel", mode: "local", id: "3f2c…" }

Parameter	Type	Description
action	string	list, get, create, update, delete; with mode: "local" also cancel, remove (a finished one), send_now (a missed or failed one)
mode	string	bot (default) or local. For list, omit to get both modes
filter	string	For list: pending (default), sent, all
id	string	Scheduled message ID (for get/update/delete/cancel/remove/send_now)
conversation_id	string	Chat to send to (for create): a person, or a group with mode: "local"
message	string	Message text. Bot: max 800 chars, no newlines. Local: up to 2000 chars
datetime	string	ISO 8601 datetime when the message should be sent
notification_type	string	Bot only: extension, whatsapp, or automatic (default)

download_media

Download and decrypt media from a message (images, videos, audio, documents).

Parameter	Type	Description
message_id	string	The message ID containing media
conversation_id	string	The conversation the message belongs to

Returns base64-encoded media data with mimetype, size, duration, and caption.

manage_notes

Read and write contact notes (WhatsApp Business).

Parameter	Type	Description
action	string	get, set
contact_id	string	The contact ID
note	string	Note text (required for set)

call_recordings

Read WhatsApp call recordings made by the Kaption extension and their transcripts, with every word timed. Each recording names its conversation (the other person, or the group for a group call). Recordings of locked chats are never returned.

Parameter	Type	Description
action	string	list, get, search
id	string	Recording ID (required for get)
search	string	Text to find in names and transcripts (required for search)
conversation_id	string	Only recordings of this contact or group
date_from / date_to	string	ISO 8601 date range of when the call started
limit	number	Max recordings (default 20, max 100)
include_words	boolean	For get: each word with its start and end in seconds

get_api_info

Get HTTP REST API connection info for programmatic access without MCP overhead. Returns URL, auth token, and available endpoints.

Multi-account support

Multiple WhatsApp accounts can be connected simultaneously (e.g. personal + business). Use target_session on any tool to route to a specific account. Query entity: "session" to see all connected accounts and their session IDs.

Multi-instance support

Multiple AI tools can share the same extension connection. The first instance starts a WebSocket hub; subsequent instances auto-detect the existing hub and relay through it. If the hub stops, a relay automatically promotes itself.

WebMCP support

Kaption is WebMCP-ready. On browsers that support the W3C WebMCP draft (navigator.modelContext, Chrome 146+), the extension automatically registers all tools with the browser's native AI tool registry. This means browser-based AI agents can discover and invoke Kaption tools without any MCP server or WebSocket connection — zero configuration.

When WebMCP is available, the extension registers tools prefixed with kaption_ (e.g. kaption_query, kaption_manage_chat) complete with JSON Schema input definitions and readOnlyHint annotations. The tools use the same handlers as the MCP server, so behavior is identical across both paths.

Security

• Localhost only — no cloud relay, no external connections
• No messages sent — AI assistants can read, organize, schedule, and draft, but never send messages directly
• Locked chats hidden — WhatsApp-locked conversations are excluded from all queries
• Feature-gated — MCP bridge must be explicitly enabled in the extension
• Business features gated — labels and notes require a WhatsApp Business account
• Rate limited — draft messages limited to 10 conversations per 5-minute window; write operations include random delays

License

BSL 1.1 — free to use, converts to MIT after 4 years.

One-time attachments (owner-only MCP_LOCAL_SEND)

Use create_attachment_upload({name,mime,size,target_session?}), PUT the file bytes to its uploadUrl within 15 minutes, then send attachment:{ref:attachmentRef} within one hour. The selected Kaption account consumes the ref once and stores it locally, so a schedule days later does not depend on the temporary upload. If a download fails after consumption, create a new upload. Maximum size: 8 MiB; image, document, audio or video.

On this local hub, attachment:{path:"/absolute/file.pdf"} uploads automatically; optional name and mime override the filename/type. Paths must name regular files. The cloud relay and WebMCP do not accept paths. Arbitrary URLs are never accepted. Existing small inline {name,mime,size,dataB64} uploads continue to work. All sends still require the extension's owner-only flag, consent and ask-first rules. No assistant can accept consent.