Quick start • Why zaileys-mcp • Install • What you can build • Configuration • Docs
接入你的智能体
發布者自己的指令。不是 PlugMCP 端點。
用我們的金鑰與日誌託管一份副本。
npx -y [email protected]Claude Code 沒有本機指令的安裝連結。複製下方指令並在終端機中執行。
claude mcp add zaileys-mcp -- npx -y zaileys-mcp@0.1.2在終端機中執行。
{ "mcpServers": { "zaileys-mcp": { "command": "npx", "args": [ "-y", "[email protected]" ] } }}claude_desktop_config.json
遠端 URL 還是本機指令? 兩種安裝方式的差異
來自發布者
Model Context Protocol server for WhatsApp, powered by Zaileys — let AI agents send messages, media, polls, and read chats from your WhatsApp. Zero-config: attach to any Zaileys client in one line, or run standalone.
[!NOTE] This README is a high-level overview. The complete guides, tool catalog, and configuration reference live in the documentation site at https://zeative.github.io/zaileys-mcp/.
[!WARNING] This drives a real WhatsApp account through the unofficial Web API (via Zaileys/Baileys). WhatsApp may suspend numbers that use unofficial automation, and an AI agent with these tools can message anyone in your account. Use
read-onlyor an allowlist when you don't need full access, and never point it at an account you can't afford to lose.
Add it to your MCP client — npx fetches it on demand. First launch prints a QR to scan in WhatsApp → Linked Devices; the session persists.
{
"mcpServers": {
"whatsapp": {
"command": "npx",
"args": ["-y", "zaileys-mcp"],
"env": { "ZAILEYS_SESSION": "my-wa" }
}
}
}
That's it. Restart your client, then ask your agent:
"Send a WhatsApp to +62 812 3456 7890 that the deploy is done ✅"
"Summarize my last 20 messages in the family group."
Works with Claude Desktop, Cursor, and any MCP client that speaks stdio.
find_tools meta-tool. Full capability, small context — no bloating every request.read-only mode and recipient allowlists for when an agent shouldn't have full reach.No install needed for the standalone server — npx handles it:
npx -y zaileys-mcp
To embed it in your own bot, add it as a dependency:
npm i zaileys-mcp zaileys # or: pnpm add • yarn add • bun add
Requires Node.js v20+. Peer dependency: zaileys >= 4.7.
60+ tools, 1:1 with the Zaileys API — every to / chat / jid accepts a phone number or a JID interchangeably.
Agent: "add +62 811 111 111 to the Weekend group and make them admin"
→ find_tools("add member to group") → enables group_add, group_promote, …
→ group_add({ group, participants })
→ group_promote({ group, participants })
send_text, send_media, send_location, send_poll, send_contact, send_sticker, react, edit_message, delete_message, forward_message, pin_message, unpin_messagelist_chats, get_messages, chat_archive/unarchive/pin/unpin/mute/unmute/mark_read/mark_unread/delete/clear, set_disappearingme, check_number, get_profile, save_contact, remove_contactsend_typing, send_recording, set_presencegroup_metadata/list/create/add/remove/promote/demote/update_subject/update_description/leave/invite_code/invite_revoke/invite_info/invite_accept/join_requests/approve_join/reject_join/setting/join_approval/member_add_mode/toggle_ephemeralcommunity_metadata/list/subgroups/create/link_group/unlink_group/leavenewsletter_metadata/messages/create/follow/unfollow/mute/unmute/reactprivacy_get, blocklist, block, unblockset_profile_name/status/picture, remove_profile_picturebusiness_profile, business_catalog, business_collections60+ tools is a lot to inject into an agent's context on every request. zaileys-mcp uses progressive tool disclosure (the default): only a small core (~13 tools) stays active, and the rest are revealed on demand via a find_tools meta-tool that enables matches and fires MCP's tools/list_changed.
ZAILEYS_TOOLS=progressive # ~13 core active, rest via find_tools (default)
ZAILEYS_TOOLS=full # all 60+ active
ZAILEYS_TOOLS=core # ~12 core, no discovery
ZAILEYS_TOOLS=send_text,get_messages # a hand-picked set (+ find_tools)
Already run a Zaileys client? Expose it to AI agents in one line — it reuses your live session:
import { Client } from 'zaileys'
import { serveMcp } from 'zaileys-mcp'
const client = new Client()
await serveMcp(client) // AI agents can now drive this WhatsApp over stdio
Read-only, allowlist, or a custom transport:
import { createMcpServer, serveMcp } from 'zaileys-mcp'
await serveMcp(client, { readOnly: true, allowlist: ['62812xxxxxxx'] })
const server = createMcpServer(client) // bring your own HTTP/SSE transport
await server.connect(myHttpTransport)
Standalone (env vars) — programmatic options mirror these one-to-one:
Two guardrails, usable in any mode: readOnly (write tools never registered) and allowlist (outbound tools refuse other recipients). Recommended: readOnly for summarizers/monitors, allowlist for personal/team notifiers, a dedicated number for full automation.
Hit a problem or have a feature request? Open an issue.
Distributed under the MIT License. See LICENSE for details.
Model Context Protocol server for WhatsApp, powered by Zaileys — let AI agents send messages, media, polls, and read chats from your WhatsApp. Zero-config: attach to any Zaileys client in one line, or
{ "mcpServers": { "zaileys-mcp": { "command": "npx", "args": [ "-y", "[email protected]" ] } }}總覽
Model Context Protocol server for WhatsApp, powered by Zaileys — let AI agents send messages, media, polls, and read chats from your WhatsApp. Zero-config: attach to any Zaileys client in one line, or run standalone.
工具
尚無工具清單。遠端伺服器會在首次健康檢查後取得工具清單。
常見問題
在本頁複製適用於 Cursor、Claude Desktop、Claude Code 或 VS Code 的 stdio 設定。它使用發布者的安裝方式,而不是 PlugMCP 代理伺服器。
這筆記錄未列出必要的環境變數。
Zaileys 透過 stdio 在本機執行。
上次檢查時間:2026-11-16。 Model Context Protocol server for WhatsApp, powered by Zaileys — let AI agents send messages, media, polls, and read chats from your WhatsApp. Zero-config: attach to any Zaileys client in one line, or run standalone.