Skip to content
MCP server

agent-in-browser

Drive Chrome, Edge, Brave or Firefox from any agent over CDP and WebDriver BiDi

Runs locallyStatus and history
Install
Pricing
Open source
Platforms
CLI, macOS, Linux
Runs
Locally
Version
0.1.2
Listing updated
today

Similar MCP servers

About agent-in-browser

agent-in-browser is an open-source MCP server for browser automation. It runs locally on your machine and is started by your AI client.

agent-in-browser is a dependency-free Go binary that gives agents a real browser without an extension or vendor runtime. Chromium-family browsers are driven through the DevTools Protocol and Firefox through WebDriver BiDi, with Safari detected but still on the roadmap. The agent reads a page as a snapshot of elements with refs, then types, clicks and fills forms by ref.

The same tool catalogue is offered three ways: an MCP server, a shell CLI that shares one persistent browser, and a scenario runner that replays YAML test steps across several browsers and writes JUnit or JSON reports. An agent skill is bundled in the binary, and setup commands write configuration for Claude Code, Cursor, Codex, Gemini CLI and VS Code.

The browser runs on its own dedicated profile, with the debugging endpoint bound to 127.0.0.1 only. The README warns that any local process can drive it, so avoid signing it into accounts you would not hand to the agent. MIT licensed.

Install

Run in a terminal

claude mcp add agent-in-browser -- npx -y https://github.com/Gwakko/agent-in-browser/releases/download/v0.1.2/agent-in-browser-0.1.2.mcpb

Add to claude_desktop_config.json

{
  "mcpServers": {
    "agent-in-browser": {
      "command": "npx",
      "args": [
        "-y",
        "https://github.com/Gwakko/agent-in-browser/releases/download/v0.1.2/agent-in-browser-0.1.2.mcpb"
      ]
    }
  }
}
Install in CursorOpens Cursor, which asks you to confirm. Or add it by hand:

Add to ~/.cursor/mcp.json

{
  "mcpServers": {
    "agent-in-browser": {
      "command": "npx",
      "args": [
        "-y",
        "https://github.com/Gwakko/agent-in-browser/releases/download/v0.1.2/agent-in-browser-0.1.2.mcpb"
      ]
    }
  }
}
Install in VS CodeOpens VS Code, which asks you to confirm. Or add it by hand:

Add to .vscode/mcp.json

{
  "servers": {
    "agent-in-browser": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "https://github.com/Gwakko/agent-in-browser/releases/download/v0.1.2/agent-in-browser-0.1.2.mcpb"
      ]
    }
  }
}

Add to ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "agent-in-browser": {
      "command": "npx",
      "args": [
        "-y",
        "https://github.com/Gwakko/agent-in-browser/releases/download/v0.1.2/agent-in-browser-0.1.2.mcpb"
      ]
    }
  }
}

Add to your client's MCP settings (most use this shape)

{
  "mcpServers": {
    "agent-in-browser": {
      "command": "npx",
      "args": [
        "-y",
        "https://github.com/Gwakko/agent-in-browser/releases/download/v0.1.2/agent-in-browser-0.1.2.mcpb"
      ]
    }
  }
}

Using several servers? Build one config file with the MCP config generator.

Tools this server provides (23)

Taken from the project documentation; the running server may differ.

browser_status
Reports browser status.
browsers_installed
Lists installed browsers.
tabs_list
Lists tabs.
tab_new
Opens a tab.
tab_select
Selects a tab.
tab_close
Closes a tab.
navigate
Navigates to a URL.
snapshot
Reads the page as element refs.
find
Finds elements.
click
Clicks an element.
hover
Hovers an element.
type
Types text.
press_key
Presses a key.
fill_form
Fills form fields.
scroll
Scrolls the page.
wait_for
Waits for text or a condition.
get_text
Gets page text.
screenshot
Takes a screenshot.
evaluate
Evaluates JavaScript.
console_messages
Reads console messages.
network_requests
Reads network requests.
resize
Resizes the viewport.
browser_close
Closes the browser.

What it does

Multi-browser support
Chromium browsers via CDP, Firefox via WebDriver BiDi.
Snapshots with refs
Pages are read as element refs the agent can act on.
Shell CLI
Call any tool from scripts, sharing one persistent browser.
Scenario runner
Run YAML tests across a browser matrix with JUnit output.
Console and network
Read console messages and network requests.
Dedicated profile
Uses its own browser profile that survives restarts.

Good for

  • Let an agent test a signup flow in a real browser
  • Run the same YAML scenario on Chrome and Firefox
  • Capture console errors while an agent browses
  • Drive a browser from shell scripts without MCP

Pricing

Open source

MIT licence

agent-in-browser FAQ

Does it need a browser extension?

No, it uses native automation protocols.

Which clients can use it?

Claude Code, Claude Desktop, Cursor, Codex, Gemini CLI, Windsurf, VS Code, Zed, opencode and other MCP clients.

Can it reuse my logged-in Chrome profile?

Not the default profile; Chrome 136 refuses that, so sign in once inside the agent's own browser window.

How do I install it?

Install script, Homebrew, go install, or a one-click .mcpb bundle for Claude Desktop.

Is agent-in-browser free?

Yes. agent-in-browser is open source, so you can use and self-host it for free.

How do I add agent-in-browser to Claude Code?

Run this command in your terminal: claude mcp add agent-in-browser -- npx -y https://github.com/Gwakko/agent-in-browser/releases/download/v0.1.2/agent-in-browser-0.1.2.mcpb

What are the best alternatives to agent-in-browser?

The closest alternatives to agent-in-browser are REA, Browser Use, Scrapling MCP Server, Context7 and Chrome DevTools MCP.

Where this comes from

  • Description: written from the project's README.
  • Install steps: from the official MCP Registry entry.

How AIPoweredBot checks servers