JSPM

  • ESM via JSPM
  • ES Module Entrypoint
  • Export Map
  • Keywords
  • License
  • Repository URL
  • TypeScript Types
  • README
  • Created
  • Published
  • Downloads 466
  • Score
    100M100P100Q112554F
  • License MIT

MCP server + CLI giving AI agents a real, stealth browser (Patchright/Playwright) — per-country identity, self-healing actions, snapshots, multi-step plans, structured extraction, CDP attach.

Package Exports

  • @fusengine/browser-mcp

Readme

fuse-browser

Give your AI agent a real, stealth browser. An MCP server + CLI on top of Patchright (stealth Chromium) with a Playwright fallback.

Your agent gets a real browser — per-country identity (locale/currency/timezone/geo), stealth fingerprint, self-healing actions, an indexed snapshot with stable refs (piercing Shadow DOM + iframes), multi-step plans, structured extraction, visual diff, and human guardrails for payments and bookings. It drives real Chromium, so it reads Next.js / SPA pages after hydration — not just static HTML.

29 MCP tools · stealth + rotating proxies · virtualized-list scraping · HAR record/replay · pixel visual-diff · human handoff.

Install

# Register with Claude Code (or any MCP client) — user scope = all projects
claude mcp add fuse-browser --scope user -- npx -y @fusengine/browser-mcp
// …or add it to your MCP config manually:
{ "mcpServers": {
  "fuse-browser": { "command": "npx", "args": ["-y", "@fusengine/browser-mcp"] }
}}

Chromium installs automatically. That's it — now just ask your agent in plain language:

"Find a hotel in Annemasse this Friday under CHF 100." "What's my Google rank for 'agence web vevey' (CH)?" "Screenshot localhost:3000 in mobile and desktop."

Prefer a terminal? Install the CLI: npm i -g @fusengine/browser-mcp

fuse-browser probe https://example.com --extract-prices
fuse-browser fetch https://books.toscrape.com/ --extract-prices   # no browser, ~10× faster

How it works

An LLM runs a perceive → decide → act loop through the tools: browser_open → browser_navigate → browser_snapshot (indexed refs + form state) → browser_act (click/fill/select/pick, returns a page diff) → browser_wait_for → browser_extract / browser_screenshot. Sensitive actions (pay / book / checkout) are blocked unless the agent passes humanApproved.

What you get

  • Stealth — Patchright neutralizes the real automation signals; per-country identity + rotating proxy pool.
  • Agentic targeting — accessibility-style snapshot with stable refs, self-healing click/fill, multi-step plans.
  • Vision (Set-of-Marks) — annotate:true on browser_snapshot/browser_act/browser_screenshot draws numbered badges (= each ref) on the page, so vision models see it and target by ref.
  • Sees everything — open Shadow DOM, same/cross-origin iframes, and virtualized/infinite lists (browser_collect).
  • Fast-path — browser_fetch impersonates a real Chrome TLS fingerprint for server-rendered HTML, no browser launch — returns clean markdown and optional contacts (extractContacts) at ~HTTP speed.
  • Data out — multi-currency prices, typed CSS extraction, contact extraction (emails/phones E.164, fastPathFirst cascade), a clean→validate→dedupe→emit pipeline, CSV export, Google SERP rank tracking.
  • Ops — persistent sessions, auto crash recovery (a crashed page is recreated in the same context and restored to its last URL between calls), storageState auto-save, HAR record/replay, pixel visual_diff, human handoff for login/2FA.

Documentation

Full reference in docs/:

Installation · CLI · MCP tools (29) · Configuration · Sessions · Extraction · Anti-bot & proxies · Library

Disclaimer

Provided as-is under MIT, no warranty. fuse-browser is a neutral, dual-use automation tool that blocks nothing by default — you alone are responsible for complying with applicable law, target sites' Terms of Service, robots.txt, and data-protection rules (GDPR, nLPD). It can honor robots.txt on request (opt-in respectRobots: true). The opt-in captcha solver and contact extraction are for authorized, lawful use only. See LEGAL.md.

License

MIT