hanzili/hanzi-browse
View on GitHub ↗let any ai agent use the local browser
174 ★52 forksJavaScriptUpdated 4mo ago
What you need to know
Context layer for browsing agents shipping 24 site playbooks as hints for the LLM, so agents finish real-world tasks on sites like X, LinkedIn, Gmail and GitHub
Install
npx hanzi-browse setup
Usage
- •Run npx hanzi-browse setup to detect agents and wire MCP
- •Give your agent tasks like unsubscribing from emails or applying to jobs; Managed mode costs $0.05/task or use BYOM free
Key features
- ✓24 site playbooks as LLM hints, not brittle scripts
- ✓Auto-detects agents like Claude Code, Cursor, Codex and installs MCP config
- ✓Browser sub-agent runs the read, plan, act, observe loop
- ✓Managed mode or free BYOM mode using your own model key
- ✓REST API and SDK for product integration
- ✓Skills like e2e-tester, social-poster and data-extractor
Best for
Users whose browsing agents fail on complex real-world websites
Caveats
- ⚠BYOM keeps data local; Managed mode processes tasks on Hanzi servers via Vertex AI
- ⚠Polyform Noncommercial license
- ⚠Requires a Chrome extension install
Platforms: Local · CloudClients: Claude Code · Cursor · Codex · Gemini CLI · VS Code · Kiro · Antigravity · OpenCode
Reviewed 2026-08-11
Topics
ai-agentai-agentsbrowser-automationchromechrome-extensionclaude-codecodexcomputer-usecursordeveloper-toolslinkedin-automationmcpmodel-context-protocolweb-automationwindsurf
- Stars
- 174★
- Forks
- 52
- Language
- JavaScript
- License
- NOASSERTION
- Created
- 2026-01-24
- Last push
- 2026-04-25