LangChain + Webfuse MCP: Browser-Controlling Research Agent

A LangGraph research agent that navigates, reads, and compares web pages through a live browser using Webfuse MCP.

How It Works

Your LangChain agent connects to a user's live browser session via Webfuse MCP. The agent can navigate, read page content, click, type, and scroll — while the user watches in real time.

Stack

• LangChain + LangGraph — ReAct agent with tool calling
• Webfuse Session MCP — 13 browser control tools via MCP
• FastAPI — SSE streaming endpoint
• Webfuse Extension — Sidebar UI for the research interface

Quick Start

Agent Server

cd agent
python -m venv venv && source venv/bin/activate
pip install -r requirements.txt

# Set your keys
export OPENAI_API_KEY=sk-...
export WEBFUSE_REST_KEY=rk_...

uvicorn agent:app --host 0.0.0.0 --port 8082

Extension

Deploy the extension/ folder to your Webfuse Space. Set the AGENT_URL environment variable to point to your agent server.

Architecture

User's Browser (Webfuse Session)
  ├── Web page (any site)
  └── Sidebar Extension
        ↕ HTTP/SSE
      Agent Server (FastAPI)
        ↕ MCP (JSON-RPC)
      Webfuse Session MCP Server
        ↕
      Browser Control (navigate, read, click, type)

Blog Post

Read the full tutorial: How to Connect LangChain to a Live Browser with Webfuse MCP

Live Demo

Try it: webfu.se/+langchain-mcp/

License

MIT