webmcp-polyfill
Enables browsers to act as MCP servers by relaying tools, resources, and prompts to AI agents via a WebSocket-to-stdio bridge.
README
webmcp-polyfill
Bridge the Model Context Protocol (MCP) into the browser. Let web pages register tools, resources, and prompts that AI agents can discover and invoke — just like a native MCP server.
Architecture
┌──────────────┐ stdio (JSON-RPC) ┌──────────────────┐ WebSocket ┌──────────────────┐
│ AI Agent │ ◄──────────────────────► │ Relay Server │ ◄───────────────► │ Browser Page │
│ (OpenCode) │ │ (src/server.ts) │ │ (BrowserMCPHost) │
└──────────────┘ │ port :3098 │ └──────────────────┘
└──────────────────┘
- Relay Server — a Bun process that speaks MCP over stdio to the AI agent, while running a WebSocket server that browser pages connect to.
- Browser Host (
BrowserMCPHost) — a WebSocket client with zero Node dependencies. Uses the native browserWebSocketAPI. - Server Host (
MCPHost) — a standalone host that listens on a TCP port for direct WebSocket connections (no relay needed). - Core Registries — pure TypeScript in-memory registries (
ToolRegistry,ResourceRegistry,PromptRegistry) shared between server and browser.
How it works
- The relay server starts as a standard MCP stdio server (
@modelcontextprotocol/sdk). - A browser page creates a
BrowserMCPHost, registers tools, and connects to the relay via WebSocket. - On connection, the relay discovers all browser-side tools via
tools/listand merges them into its own registry. - The AI agent sees a unified tool list. When it invokes a browser tool, the relay proxies the
tools/callover WebSocket to the browser. - When the browser disconnects, its tools are automatically cleaned up.
Relationship to WebMCP / MCP
This project polyfills MCP into environments that can't run a TCP server — specifically browsers. Under the standard MCP model, a tool host must listen on a socket or pipe. A browser tab can't do that. webmcp-polyfill inverts the connection: the relay listens, and browsers connect as clients. The relay then translates between the agent's stdio MCP transport and the browser's WebSocket transport.
The webmcp_* prefixed tools (webmcp_resources_list, webmcp_resources_read, webmcp_prompts_list, webmcp_prompts_get) are bridge tools that expose resources and prompts through the MCP tool interface, since MCP stdio clients primarily support the tools/ namespace.
Use Cases
- Web-based devtools — register browser extension tools (DOM inspection, performance profiling, screenshot capture) that AI agents can call directly.
- In-browser automation — let an agent control a web app by invoking tools that click buttons, fill forms, or navigate pages.
- Live dashboards — expose browser-side data (WebSocket feeds, canvas state, WebGL stats) as MCP resources readable by agents.
- Hybrid workflows — combine server-side tools (file system, database) with browser-side tools (page state, user interactions) in a single session.
- Prototyping MCP servers — use the browser as a quick REPL to register and test tools without setting up a Node/Bun project.
Quick Start
# Install dependencies
bun install
# Generate self-signed TLS certs (for WSS support)
bun run gen-cert
# Start the relay server (MCP over stdio)
bun run mcp
# In another terminal, start the browser demo
bun run web
Minimal browser host
import { BrowserMCPHost } from "webmcp-polyfill/sdk/browser";
const host = new BrowserMCPHost("ws://localhost:3098");
host.tool("greet", {
description: "Greet someone by name",
inputSchema: {
type: "object",
properties: { name: { type: "string" } },
required: ["name"],
},
execute: async (args) => {
return { greeting: `Hello, ${args.name}!` };
},
});
host.connect();
Minimal server host
import { MCPHost } from "webmcp-polyfill/sdk";
const host = new MCPHost({ port: 3098 });
host.tool("add", {
description: "Add two numbers",
inputSchema: {
type: "object",
properties: { a: { type: "number" }, b: { type: "number" } },
required: ["a", "b"],
},
execute: async ({ a, b }) => a + b,
});
host.listen();
License
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.