pons
Bridge WhatsApp Business Cloud API to MCP-compatible clients, enabling AI assistants to read, search, and send WhatsApp messages.
README
<div align="center">
<br />
┌─────────────────────────────────────────┐
│ │
│ ██████╗ ██████╗ ███╗ ██╗███████╗ │
│ ██╔══██╗██╔═══██╗████╗ ██║██╔════╝ │
│ ██████╔╝██║ ██║██╔██╗ ██║███████╗ │
│ ██╔═══╝ ██║ ██║██║╚██╗██║╚════██║ │
│ ██║ ╚██████╔╝██║ ╚████║███████║ │
│ ╚═╝ ╚═════╝ ╚═╝ ╚═══╝╚══════╝ │
│ │
└─────────────────────────────────────────┘
WhatsApp in your terminal. Messages in your AI.
Bridge the WhatsApp Business Cloud API to any MCP-compatible client.<br /> Send and receive messages from Claude, Cursor, or your own tools.
pons.chat · Docs · Report Bug · Request Feature
<br />
</div>
The Problem
WhatsApp Business has a powerful API, but it's locked behind REST calls, webhook plumbing, and token management. There's no way to plug it into the tools you actually use — AI assistants, dev environments, automation pipelines.
Pons (Latin for "bridge") connects WhatsApp to the Model Context Protocol, so your AI assistant can read, search, and send WhatsApp messages as naturally as it reads files or searches the web.
What It Does
- Full WhatsApp inbox — real-time web UI with conversations, media, delivery receipts
- MCP server — expose WhatsApp as tools for Claude Desktop, Cursor, or any MCP client
- Multi-tenant — multiple WhatsApp Business Accounts, multiple users per account
- Media handling — images, videos, documents auto-downloaded to Convex storage (Meta URLs expire in 5 min)
- 24h window tracking — know when you can send free-form vs. template-only messages
- API key management — scoped keys for different clients with expiration
Tech Stack
| Layer | Choice |
|---|---|
| Framework | Next.js 16 (App Router, Turbopack) |
| Backend | Convex (database, auth, file storage, scheduled actions) |
| Auth | Convex Auth (Google OAuth) |
| MCP | @modelcontextprotocol/sdk — Streamable HTTP transport |
| UI | shadcn/ui + Tailwind CSS v4 |
| Domain | pons.chat |
| Hosting | Vercel (FRA1) + Convex Cloud (eu-west-1) |
Quick Start
Prerequisites
- Node.js 18+, pnpm
- A Meta Business App with WhatsApp Cloud API access
1. Install & configure
git clone https://github.com/NicolaiSchmid/pons.git
cd pons
pnpm install
# Start Convex dev (opens browser for auth on first run)
npx convex dev
# In another terminal
pnpm run dev:next
2. Connect WhatsApp
- Create a Meta Business App → add WhatsApp product → get test phone number
- Sign in to Pons at
http://localhost:3000 - Create an Account with your WABA ID, Phone Number ID, Access Token, and App Secret
- Set webhook URL in Meta to
https://your-domain.com/api/webhook - Use the generated Webhook Verify Token to verify
3. Connect your AI
Go to API Keys in the dashboard, create a key, and add to your MCP config:
{
"mcpServers": {
"pons": {
"url": "https://your-pons-domain.com/api/mcp",
"headers": {
"Authorization": "Bearer pons_your_api_key_here"
}
}
}
}
MCP Tools
| Tool | Scope | Description |
|---|---|---|
list_conversations |
read | List recent conversations with preview |
get_conversation |
read | Get full conversation with messages |
search_messages |
read | Search messages by content |
send_text |
send | Send a text message |
send_template |
send | Send a template message (for closed windows) |
list_templates |
read | List available message templates |
mark_as_read |
write | Mark conversation as read |
send_reaction |
write | React to a message with emoji |
Scopes: read (view), write (mark read, react), send (send messages). Assign per key.
Architecture
┌──────────────┐ ┌──────────────┐ ┌──────────────┐
│ │ │ │ │ │
│ Claude / │────▶│ Pons MCP │────▶│ Convex │
│ Cursor │ MCP │ Endpoint │ │ Backend │
│ │ │ (Next.js) │ │ │
└──────────────┘ └──────┬───────┘ └──────┬───────┘
│ │
┌──────┴───────┐ ┌──────┴───────┐
│ │ │ │
│ Pons Web │ │ Meta Graph │
│ Dashboard │ │ API v22.0 │
│ │ │ │
└──────────────┘ └──────────────┘
Data Model
Account (WhatsApp Business Account)
├── AccountMember (owner / admin / member)
├── Contact (customer phone numbers)
│ └── Conversation (thread with a contact)
│ └── Message (text, media, location, reaction...)
├── Template (pre-approved message templates)
├── ApiKey (scoped MCP authentication)
└── WebhookLog (raw payloads for debugging)
Project Structure
pons/
├── convex/ # Backend
│ ├── schema.ts # Database schema
│ ├── auth.ts # Convex Auth (Google OAuth)
│ ├── accounts.ts # Account + member management
│ ├── conversations.ts # Conversation queries
│ ├── messages.ts # Message CRUD
│ ├── mcp.ts # MCP queries + API key management
│ ├── mcpNode.ts # Crypto operations (Node.js runtime)
│ ├── webhook.ts # Webhook ingestion + media download
│ └── whatsapp.ts # Meta Graph API actions
├── src/
│ ├── app/
│ │ ├── api/mcp/ # MCP HTTP endpoint
│ │ ├── api/webhook/ # WhatsApp webhook handler
│ │ └── page.tsx # Landing + auth gate
│ ├── components/
│ │ ├── Dashboard.tsx # Main shell
│ │ ├── ConversationList.tsx # Sidebar
│ │ ├── MessageThread.tsx # Chat view
│ │ ├── AccountSettings.tsx # Settings + member management
│ │ ├── ApiKeyManager.tsx # API key CRUD
│ │ ├── AccountSelector.tsx # Account switcher
│ │ └── SetupAccount.tsx # Onboarding
│ └── components/ui/ # shadcn/ui primitives
├── middleware.ts # Auth middleware (webhook excluded)
└── vercel.json # Vercel config (FRA1 region)
Development
pnpm dev # Convex + Next.js in parallel
pnpm run check:write # Biome lint + format (auto-fix)
pnpm run typecheck # TypeScript check
pnpm run build # Production build
Deployment
- Connect GitHub repo to Vercel
- Set environment variables:
NEXT_PUBLIC_CONVEX_URL,CONVEX_DEPLOY_KEY - Build command is pre-configured:
pnpm convex deploy --cmd 'pnpm run build' - Disable Vercel Deployment Protection for production (webhook needs unauthenticated access)
License
MIT
Built by nicolaischmid.com
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。