Webmcp, developed by Jasonjmcghee, is an in-page Model Context Protocol (MCP) server for the mcp platform. This free JavaScript library allows web applications to expose client-side logic and DOM interaction to Large Language Models or AI agents. It transforms websites into tool repositories and context providers, enabling AI to interact directly with web content. Webmcp aims to bridge the gap between AI capabilities and live web environments, enhancing collaborative workflows between users and agents within web interfaces.

What does Webmcp connect to?

Webmcp connects web applications to Large Language Models (LLMs) or AI agents, serving as an in-page Model Context Protocol (MCP) server. It exposes client-side logic and DOM interaction, allowing web applications to function as tool repositories and context providers for these agents. Communication primarily uses a localhost WebSocket server to bridge between an AI host and the browser tab. The broader WebMCP standard supports additional transport protocols including stdio, Server-Sent Events (SSE), and Streamable HTTP. Compatible AI clients include Claude Desktop and VS Code.

Core features and tools

Webmcp functions as an in-page MCP server, exposing client-side logic and DOM interaction to connected AI agents. It transforms web applications into tool repositories and context providers. A core feature is its built-in Token generator tool, which end-users use to create tokens for connecting to WebMCP websites. This design avoids sharing API keys. The system operates within the user's live session, using existing cookies, session information, and page context, often removing the need for separate logins. Websites are expected to apply their own access controls.

Who is it for?

Webmcp is for developers who want to expose web application functionality to AI agents. It supports collaborative workflows where users and agents interact within the same web interface. This includes enabling AI agents to actuate web content designed for human interaction. It also suits scenarios where a user and browser agent collaborate, such as filtering templates or making visual edits on a design website. Skip this if your application does not require AI agents to interact directly with client-side web content.

How to get started

The system is designed to avoid sharing API keys, with end-users generating tokens via the built-in Token generator tool for connection. Communication primarily uses a localhost WebSocket server.

Use cases

• Enabling AI agents to actuate web content designed for human interaction.
• Facilitating collaborative tasks, such as a user interacting with a browser agent to filter templates and make visual edits on a design website.
• Allowing agents to fill structured forms (e.g., a submit_application tool) or perform quick application debugging (e.g., a run_diagnostics tool).

Webmcp by Jasonjmcghee offers a method for integrating AI agents with web applications, turning websites into interactive tools for LLMs. Its design prioritizes secure connections through user-generated tokens and operates within existing user sessions. This allows for dynamic collaboration between humans and AI directly within web interfaces. Try enabling an AI agent to fill a structured form using the submit_application tool.