WebMCP Agent

A browser assistant that reads the current page, takes files and images, and uses the page's WebMCP tools with your own model.

WebMCP Agent puts an AI assistant in Chrome's side panel that answers questions about the page you are on, using a model you connect yourself. It sends the page's title, URL, and visible text with your message to any OpenAI-compatible Chat Completions endpoint — the built-in default is DeepSeek's deepseek-flash — and you can attach up to four images and four text or code files per message. On pages that expose WebMCP tools, the model can look things up and act through them; anything that changes the page asks for your approval first, and a built-in inspector shows which tools a page actually provides and why it has none. Saved skills turn repeated instructions into entries you pick by typing /, and they import and export as SKILL.md. Mermaid diagrams in answers are drawn inline, enlarged, or saved as SVG or PNG. The conversation can move into a floating window, and the interface follows Chrome's light and dark themes in English or Simplified Chinese. There is no server of its own: no accounts, no analytics, no advertising. Your API key and conversation stay on your device and are erased when Chrome closes.