LM Studio + Browser MCP: local browser automation setup

This tutorial connects a local LLM in LM Studio to Browser MCP so an AI agent can interact with web pages. The setup requires Node.js and the Browser MCP Chrome extension, which the presenter says also works in most Chromium-based browsers. He copies the extension's server configuration into LM Studio's mcp.json, loads a model, and connects the extension.

Model selection matters. The presenter recommends updating LM Studio's llama.cpp runtime and choosing a model with tool calling support, marked by a hammer icon. He favors the largest model that runs comfortably on the available hardware and suggests 4-bit quantization. Reasoning support is optional. The example uses Gemma 4 with 4 billion parameters and a context length of roughly 50,000 to 60,000 tokens to accommodate tool output and reasoning.

The browser task asks the model to find Gemini AI's account on x.com and report its latest post. Several attempts fail because the model does not recognize the Browser MCP connection. The presenter retries by reconnecting the extension, reloading the model, and starting a new chat. He later reports that the model opens a Gemini post, though it also clicks unintended controls. He suggests using a larger browser window and testing models against hardware limits. Email replies are a proposed use, not a demonstrated result, and he recommends monitoring browser actions.