My agents kept struggling to use the web reliably.
I tried Playwright MCP, Chrome DevTools, browser-use, and most browser tools out there, but they either exposed too much low-level state or flooded the context with noisy DOM representations and accessibility trees.
So I built agent-web-interface: a browser interface for agents that represents the page as a clean, structured domain model, exposes deterministic tools for navigation and action, and keeps token usage low enough for agents to reason instead of drowning in context.
Install it in Claude Code with: claude mcp add agent-web-interface -- npx agent-web-interface@latest
Top comments (0)