About Hanzi Browse
Hanzi Browse gives an AI coding agent a real browser to work with. It installs a Chrome extension and an MCP server, then lets agents such as Claude Code, Cursor and Codex delegate browser work like clicking, reading pages and filling forms in the user's own signed in Chrome. A browser sub agent runs the loop of reading the page, planning, acting and observing, then returns a clean answer so the main agent keeps its context free for code.
It ships 24 site playbooks, which are hints for the language model rather than brittle scripts, covering sites such as X, LinkedIn, Gmail, GitHub, Notion, Figma, Slack, Reddit, Amazon and Zillow. Setup is a single command, npx hanzi browse setup, with open source skills for workflows like end to end testing. Product teams can also use a REST API and an SDK to send a plain English task that runs in their own users' browsers, with users pairing their own Chrome so credentials are never shared.
Pricing has two modes. With the managed option the vendor handles the AI, with 20 free tasks per month and then $0.05 per completed task, and errors and timeouts are free. With the bring your own model option, users supply a Claude Pro or GPT Plus subscription or any API key, and the tool is free and runs locally. No separate pricing page is published.