browser-pilot

日本語版 README

An MCP server for controlling Chrome browsers from Claude Code. Operates real browsers via a Chrome Extension.

Quick Start

1. Run setup

npx @youtyan/browser-pilot setup

This will:

• Detect your MCP host (Claude Code / Cursor / Windsurf) and generate config
• Install skills
• Guide you through Chrome Extension installation

2. Load the Chrome Extension

Open chrome://extensions, enable Developer Mode, click "Load unpacked", and select the extension path shown by setup.

3. Verify connection

npx @youtyan/browser-pilot doctor

No token required — the Extension connects automatically.

Skills

Seven skills are included:

| Skill | Description | |---|---| | bp-usage | Browser operation guide and tool quick reference | | bp-testing | Agent-driven web app testing | | bp-test-scripts | Code-driven browser testing (HTTP API) | | bp-gemini-image | Image generation via Gemini Web UI | | bp-x-operation | X (Twitter) operations (post, search, collect) | | bp-annotate-coords | Screenshot annotation with coordinates and bounds | | bp-generate-manual | Site operation manual generation with screenshots/video |

You can also install skills with npx skills add:

npx skills add https://github.com/youtyan/browser-pilot

Available Tools (19)

Core (9)

| Tool | Description | |---|---| | browser_state | Page content (mode: page) or element search (mode: find). Replaces browser_get_page + browser_find_elements | | browser_click | Click by text, CSS selector, ref, or backendNodeId | | browser_type | Set form values, select options, press keys (action: input/select/key) | | browser_scroll | Scroll page or element | | browser_wait | Wait for element appear/disappear/count change, or simple ms delay. assert param for error on timeout | | browser_batch | Execute multiple actions in one round-trip | | browser_tab | Tab management + navigation (list, connect, create, navigate, back, forward, reload, action log, manual/test generation) | | browser_upload | Upload or paste files | | browser_screenshot | Take screenshot (full page or element, with savePath) |

Debug (2)

| Tool | Description | |---|---| | browser_debug | Console, network, performance, health, CSS inspect, interaction diagnosis (action parameter) | | browser_eval | Execute JavaScript in page context (CDP direct, fast). First-class tool, no env var gate |

Advanced (3)

| Tool | Description | |---|---| | browser_mouse | Mouse actions (hover, dblclick, drag, contextmenu) | | browser_extract | Schema-first structured data extraction | | browser_artifacts | Record video, save PDF, annotate screenshots (action: record/pdf/annotate) |

Storage (1)

| Tool | Description | |---|---| | browser_storage | Cookie, localStorage, sessionStorage management (target + action params) |

Integrations (4)

| Tool | Description | |---|---| | gemini_generate_image | Generate images via Gemini Web UI | | x_collect_tweets | Collect tweets from X (Twitter) with scroll and dedup | | x_post | Post to X (Twitter) | | x_interact | Batch interact on X (like, repost, follow) |

Security Considerations

browser-pilot is a powerful browser automation tool. Understand the following risks before use.

• browser_eval executes arbitrary JavaScript in the MAIN world of Chrome tabs. This gives full access to the page's DOM, cookies, localStorage, and more.
• The MCP server binds to localhost:18888. Any local process can connect (no token required for WebSocket).
• The HTTP API (/api/*) is gated by a bearer token stored in ~/.browser-pilot-token with 0600 permissions.
• The Chrome Extension has <all_urls> permission. It can access any website.
• Only use this tool in trusted environments.

Development

See DEVELOPMENT.md for setup, architecture, adding tools, and build instructions.

Updating

npx @youtyan/browser-pilot update

License

MIT