Charlotte

概要

Charlotte is an MCP server that makes the web readable for AI agents. It intelligently renders web pages into structured, agent-readable representations using headless Chromium, exposing the browser's semantic understanding to agents via Model Context Protocol tools. Unlike other solutions that dump large, unstructured accessibility trees, Charlotte decomposes pages into typed structures and offers multiple detail levels, significantly reducing token usage and enabling precise navigation, observation, and interaction without relying on vision models or brittle selectors.

主な機能

• Comprehensive navigation controls including `navigate`, `back`, `forward`, and `reload`
• Development mode features including `dev_serve` for static serving with auto-reload, `dev_inject` for CSS/JS, and `dev_audit` for accessibility and performance checks
• Advanced observation capabilities like `observe`, semantic and spatial `find`, `screenshot`, and structural `diff`
• Extensive interaction tools for `click`, `type`, `select`, `toggle`, `submit`, `scroll`, `hover`, and more
• Efficient, structured web page representation with configurable detail levels (minimal, summary, full)
• 73 GitHub stars

ユースケース

• Automate complex web interactions such as filling out forms and interacting with dynamic elements.
• Enable AI agents to efficiently browse and orient themselves on web pages with minimal token consumption.
• Provide a local development feedback loop with live reloading, CSS/JS injection, and web audits for web projects.