browsermcp.ai

Give your AI a browser it can use. One command. No cloud. No API key. No cost.

browsermcp lets AI tools like Claude Code, Cursor, and Codex CLI browse the web, read pages, fill out forms, take screenshots, and track changes — all running privately on your machine.

What can it do?

| You say to your AI... | What happens | |------------------------|-------------| | "Check the Stripe docs for breaking changes" | browse — opens the page, strips the clutter, gives your AI clean text instead of raw HTML | | "Get the pricing tiers from Linear's website" | extract — reads the page and structures the data however you need it | | "Screenshot our homepage after that deploy" | screenshot — captures a full-page or element-level PNG | | "Has anything changed on the AWS status page?" | watch — compares the page to last time and only shows what's different | | "Test our signup form with a dummy account" | interact — fills inputs, clicks buttons, submits forms |

Setup (30 seconds)

Step 1: Run this in your terminal:

npx browsermcpai init

Step 2: It prints a config block. Copy it into your AI tool's settings:

• Claude Code — paste into .mcp.json in your project folder
• Cursor — paste into ~/.cursor/mcp.json
• Codex CLI — paste into ~/.codex/config.json

Step 3: Restart your AI tool. Done — your AI can now browse the web.

Not sure where to paste it? Run npx browsermcpai init — it detects your tools and tells you exactly where.

How it works

Your AI (Claude Code / Cursor / Codex CLI)
  |
  |  asks browsermcp to browse a page
  v
browsermcp (runs on your machine)
  |
  |  opens a real browser, reads the page,
  |  strips out all the junk (ads, scripts, menus),
  |  and returns just the useful content
  v
Your AI gets clean, readable text

Why this matters: A typical webpage is 80,000+ tokens as raw HTML. browsermcp distills it down to 1,000-8,000 tokens — 10 to 50x smaller. Your AI reads faster, understands better, and costs less.

The five tools

browse

Read any webpage. Returns clean markdown text.

browse({ url: "https://example.com/pricing" })

Add an instruction to focus on what matters:

browse({ url: "https://example.com/pricing", instruction: "just the plan names and prices" })

extract

Same as browse, but you tell it what shape you want the data in:

extract({
  url: "https://example.com/team",
  schema: { people: [{ name: "string", role: "string" }] }
})

screenshot

Capture what a page looks like:

screenshot({ url: "https://myapp.dev", fullPage: true })

Or just one part of the page:

screenshot({ url: "https://myapp.dev", selector: "#hero" })

watch

Track changes on a page over time:

watch({ url: "https://docs.stripe.com/changelog" })

First time: saves a snapshot and returns the full page. Next time: returns only what changed — 90%+ fewer tokens.

interact

Fill forms, click buttons, test flows:

interact({
  url: "https://myapp.dev/signup",
  actions: [
    { type: "fill", selector: "input[name='email']", value: "[email protected]" },
    { type: "click", selector: "button[type='submit']" }
  ]
})

Staying in control

browsermcp tracks how much your AI browses and stops it if it goes overboard.

npx browsermcpai usage    # see today's and weekly stats

Default limits (changeable in .browsermcp.json):

• 100 browsing sessions per day
• 1 million tokens per day

How does this compare?

| | browsermcp | Browserbase | Browser-Use | Stagehand | |--|------------|-------------|-------------|-----------| | Setup | One command | Sign up + API key | pip install + config | npm install + API key | | Cost | Free | $20-99/mo | Free (self-host) | Free + Browserbase | | Runs on | Your machine | Cloud | Your machine | Cloud (optional) | | Token savings | 10-50x built-in | None | Some | Some | | Works with | Claude Code, Cursor, Codex CLI | Custom SDKs | Python agents | JS agents |

Configuration

npx browsermcpai init creates a .browsermcp.json file in your project:

{
  "browser": {
    "timeout": 30000,
    "headless": true
  },
  "distill": {
    "maxTokens": 4000,
    "includeLinks": true,
    "includeImages": false
  },
  "limits": {
    "maxSessionsPerDay": 100,
    "maxTokensPerDay": 1000000
  }
}

You can also put a config at ~/.config/browsermcp/config.json for global defaults. Project-level config takes priority.

Requirements

• Node.js 20 or later
• That's it. Playwright installs its own browser automatically.

License

MIT