The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
Hanzi Browse
Let any ai agent use the local browser
Install
npx hanzi-browse setupInstall command taken from the project's README or registry entry. Always review code before running it.
README excerpt
Read full README on GitHub →Hanzi Browse
The context layer for browsing agents.
Your browsing agent keeps failing on real sites — X uses Draft.js, LinkedIn hides the connect button, Gmail needs keyboard shortcuts. Hanzi Browse ships 24 site playbooks — hints for the LLM, not brittle scripts — so it actually finishes the task.
Works with
Two ways to use Hanzi Browse
Same 24 site playbooks underneath. Two install paths depending on who's driving.
For your agent — a browser sub-agent for your coding agent
One command. npx hanzi-browse setup detects every AI agent on your machine (Claude Code, Cursor, Codex, and 9 more) and wires Hanzi Browse in as an MCP tool. Your main agent delegates browser work; a sub-agent runs the loop — read page → plan next action → click/type/scroll → observe → repeat until done — and returns a clean answer. Site playbooks auto-load by URL so the model already knows the quirks.
For your product — browser automation for your users, described in English
Your backend calls runTask({ task: "…" }). Your users' own Chrome executes it, signed in as themselves. Same 24 playbooks as the CLI, exposed as a REST API and @hanzi-browse/sdk. Free tools on tools.hanzilla.co are built on this SDK.
Get Started
npx hanzi-browse setup
One command does everything:
npx hanzi-browse setup
│
├── 1. Detect browsers ──── Chrome, Brave, Edge, Arc, Chromium
│
├── 2. Install extension ── Opens Chrome Web Store, waits for install
│
├── 3. Detect AI agents ─── Claude Code, Cursor, Codex, Windsurf,
│ VS Code, Gemini CLI, Amp, Cline, Roo Code
│
├── 4. Configure MCP ────── Merges hanzi-browse into each agent's config
│
├── 5. Install skills ───── Copies browser skills into each agent
│
└── 6. Choose AI mode ───── Managed ($0.05/task) or BYOM (free forever)
- Managed — we handle the AI. 20 free tasks/month, then $0.05/task. No API key needed.
- BYOM — use your Claude Pro/Max subscription, GPT Plus, or any API key. Free forever, runs locally.
Examples
"Go to Gmail and unsubscribe from all marketing emails from the last week"
"Apply for the senior engineer position on careers.acme.com"
"Log into my bank and download last month's statement"
"Find AI engineer jobs on LinkedIn in San Francisco"
Skills & Free Tools
Hanzi Browse has two distribution channels. Both use the same browser automation engine and site domain knowledge:
Skills — for users who run Hanzi Browse locally through their AI agent. The setup wizard installs skills directly into your agent (Claude Code, Cursor, etc.). Each skill teaches the agent when and how to use the browser for a specific workflow.
Free Tools — hosted web apps that anyone can try without installing anything. Each tool is a standalone app built on the Hanzi Browse API that demonstrates a use case. Every skill can become a free tool.
Skills
Installed automatically during npx hanzi-browse setup. Your agent reads these as markdown files.
Similar plugs
Reverse engineer anything with agents, from app behavior down to native binaries.
Build Agentic workflows, RAG pipelines, with rich AI model and tool support on one collaborative workspace. Deploy on cloud, VPC, or self-hosted, so teams move from prototype…
User-friendly AI Interface (Supports Ollama, OpenAI API, ...)
A cross-platform desktop All-in-One assistant for Claude Code, Codex, OpenCode, OpenClaw, Grok Build & Hermes Agent. Only official website: ccswitch.io