WebMCP
TypeScript SDK for Web AI Agents
Bi-directional communication between AI agent runtimes and browser content via Model Context Protocol.
Quick Start
import { WebMCPServer, WebStandardHttpTransport } from '@thestudioxi/webmcp';
const server = new WebMCPServer({
transport: new WebStandardHttpTransport(),
});
server.registerTool(
{
name: 'get_page_content',
description: 'Extract content from the active browser tab',
inputSchema: {
type: 'object',
properties: {
selector: { type: 'string', description: 'CSS selector' },
},
},
},
async (args) => {
// AI agent can now read browser DOM in real time
return { content: document.querySelector(args.selector)?.textContent };
}
);
await server.start();Installation
bun add @thestudioxi/webmcpnpm install @thestudioxi/webmcppnpm add @thestudioxi/webmcpFeatures
Browser-Native Communication
Inspect and manipulate active browser tab DOMs in real time without heavy headless browsers. Built on WebSocket, MessageChannel, and HTTP transports.
Zero-Config Tool Registration
Register and call web tools via bi-directional transports using a standard JSON-RPC protocol. Works with any framework.
Framework Adapters
First-class adapters for Express.js, Hono, NestJS, and Vercel AI SDK. Zero framework-specific dependencies in the core.
AI-Native Architecture
Dual SDK design with WebMCPBrowserBridge (frontend) and WebMCPAgentClient (backend). Built for autonomous agent loops.
Type-Safe & Modular
Full TypeScript declarations, transport-agnostic interfaces, and modular architecture. No implicit any types.
Lightweight & Fast
Built on Bun with sub-10ms JSON serialization overhead. Zero-dependency adapters that wrap the Web Standard Request/Response API.
Framework Adapters
First-class integrations — zero framework-specific dependencies in the core SDK.