Mcpbridge – Firefox MCP Bridge for Claude

McPbridge is a lightweight server that exposes the Model Context Protocol (MCP) interface for the Claude web application, enabling it to communicate directly with the Firefox browser. The primary problem it solves is the lack of a native, bidirectional bridge between an AI assistant running in a web context and the browser’s internal APIs. By turning Firefox into a fully‑featured MCP server, developers can harness Claude’s capabilities to manipulate the browsing session, retrieve page data, and trigger custom actions without leaving the browser environment.

At its core, McPbridge listens for MCP requests over a local socket and forwards them to the Firefox instance via its remote debugging protocol. This allows Claude to invoke browser tools such as opening new tabs, capturing screenshots, or extracting DOM elements. For developers, this means they can write AI‑driven workflows that interact seamlessly with web pages—e.g., “summarize the article on this page” or “fill out and submit the form below”—and receive structured responses back in Claude’s conversational interface.

Key features include:

• Resource Exposure: The bridge presents a set of browser resources (tabs, windows, network requests) as MCP resources that can be queried or modified.
• Tool Integration: Custom tools such as , , and are made available to Claude, enabling rich interactions.
• Prompt Management: Pre‑defined prompts can be loaded into the MCP server, allowing quick invocation of common browsing tasks.
• Sampling Controls: The server supports fine‑grained control over text generation sampling, letting developers tweak Claude’s output directly from the browser context.

Typical use cases span automated testing, content scraping, and interactive tutorials. For example, a QA engineer could ask Claude to “open the login page, fill in credentials, and verify the success message,” while a content creator might request “extract all headings from this article and format them as Markdown.” In research settings, McPbridge allows rapid prototyping of AI‑powered web crawlers that adapt their behavior based on Claude’s reasoning.

Integration into existing AI workflows is straightforward: a developer embeds the MCP server in their local environment, configures Claude’s web client to point at the bridge’s endpoint, and then orchestrates complex browser interactions through natural language commands. The server’s lightweight design ensures minimal overhead, and its dual MIT/Apache 2.0 licensing encourages community contributions while maintaining clear legal boundaries.

Overall, McPbridge transforms Firefox into a programmable interface for Claude, bridging the gap between AI assistants and web browsers. It empowers developers to build sophisticated, conversationally driven browsing experiences without writing custom browser extensions or relying on third‑party services.