WebMCP gives AI agents control of your real browser — your browser, your data, your machine. No cloud sandboxes, no proxies.

It connects AI agents (Claude, Cursor, custom LLMs) to a Chrome extension via the Model Context Protocol (MCP), letting them navigate, click, type, and read pages exactly like you would.

• Real browser — agents use your actual Chrome with your cookies, sessions, and logins
• MCP-native — built on the open standard for AI tool communication
• Privacy-first — everything runs locally, nothing leaves your machine
• 25 automation tools — navigate, read pages, fill forms, manage tabs, and more
• Accessibility tree — agents see pages as structured elements with reference numbers, not raw HTML
• Human oversight — monitoring popup, kill switch, CAPTCHA detection

Install from the Chrome Web Store, or load unpacked from webagent-extension/ via chrome://extensions.

cd webagent-mcp
npm install
npm start          # stdio transport (for Claude, Cursor)
# or
npm run dev        # SSE transport on port 3000

Configure your AI tool to use the WebMCP MCP server. The extension connects automatically via WebSocket.

• CLAUDE.md — Project index for AI agents
• ARCHITECTURE.md — System design and data flow
• Integration Guide — SDK usage and examples
• Protocol Spec — Message formats and actions
• Changelog — Version history

WebMCP requires 5 Chrome permissions and 1 host permission:

Website: webmcp.tanujmittal.com