Browserful

Your AI Agent

Claude, Cursor, or any MCP client

→

MCP Server

→

Web Browser

Extension or headless

→

Any Website

Returns structured data

How It Works

Four simple operations

Your agent sends commands, Browserful executes them and returns clean, structured data.

navigate

Go to URL

Opens a page and returns title, URL, text content, and all links.

click

Click Element

Clicks buttons, links, or any element by CSS selector.

type

Enter Text

Types into input fields, search boxes, and forms.

End Session

Closes the browser session and frees resources.

Structured output your agent can use

Every action returns clean JSON with the page state. No HTML parsing, no guessing—just the data your agent needs to decide what to do next.

Page title and current URL
Extracted text content
All links with labels and URLs
Works with any LLM

Response from navigate("https://example.com")

{
  "title": "Example Domain",
  "url": "https://example.com",
  "text": "This domain is for use in illustrative examples in documents...",
  "links": [
    {
      "text": "More information...",
      "href": "https://www.iana.org/domains/example"
    }
  ]
}

Browser Backends

Your browser or ours

Same API, different browser sources. Pick what fits your use case.

🔌

Extension

Control your actual browser with your existing logins

Uses your cookies and sessions
Access authenticated pages
Watch actions in real-time
Works with bot-protected sites

Example: "Log into my email and summarize unread messages"

⚡

Engine

Spawn fresh headless browser instances

Clean browser, no existing state
Run headless or visible
Scale to hundreds in parallel
Deploy locally or in cloud

Example: "Scrape product prices from 100 pages"

Output Modes

Text now, visual soon

Text mode works with any LLM today. Visual mode coming for multimodal models.

Available Now

Text Mode

Structured data that works with any language model

Page title, URL, and metadata
Extracted text content
All links with labels
Works with Claude, GPT, Llama, etc.

Coming Soon

Visual Mode

Screenshots for multimodal language models

Viewport screenshots as images
MCP image content blocks
See layouts, charts, images
Requires Claude 3, GPT-4V, etc.

Deploy

Run it your way

Local development, self-hosted, or managed cloud.

Local

Free forever

Run on your machine with your browser or local headless engine.

git clone github.com/contextbrowser/contextbrowser
cd contextbrowser
go build && ./context-browser

Self-Hosted

Free, open source

Deploy the full stack to your own infrastructure.

git clone github.com/contextbrowser/contextbrowser
cd contextbrowser
docker-compose up

Cloud

Coming soon

We run the infrastructure. You just connect your agent.

{
  "mcpServers": {
    "browser": {
      "url": "mcp.browserful.com"
    }
  }
}

Web access for AI agents