Skip to main content
The MCP-B browser extension bridges your browser and desktop AI. It collects WebMCP tools from all your open tabs and bridges them to local MCP clients like Claude Desktop. Additionally, it includes specialized AI agents that help you build userscripts and create new WebMCP tools without leaving the browser.

Install MCP-B Extension

Get the extension from the Chrome Web Store

​
Connect to Desktop AI with Native Host

Unlock the full power of WebMCP: The native host bridges your browser tools to desktop AI assistants like Claude Code and Claude Desktop. This means you can use browser-based tools from your command line or desktop apps!

Setup Native Host

Start here - Connect browser tools to Claude Code, Claude Desktop, and other MCP clients

How It Works

Learn about the architecture behind browser-to-desktop integration

​
Quick Navigation

Jump to the section that best matches your needs:

Understanding Agents

Learn about the 4 specialized agents and when to use each one

Managing Userscripts

Create, download, upload, and organize custom scripts

​
What Can You Do?

Connect Desktop AI

Use browser tools from Claude Code, Claude Desktop, and other desktop MCP clients

AI-Powered Development

Work with specialized AI agents that understand web development and can build scripts for you

Customize Websites

Add dark modes, remove ads, change layouts, and enhance any website with custom userscripts

Automate Tasks

Auto-fill forms, extract data, and streamline repetitive workflows with intelligent scripts

Build MCP Tools

Expose website functionality as tools that AI agents can discover and use

Cross-Site Workflows

Chain actions across multiple tabs and sites for complex automation

​
Quick Start

1

Install the extension

Download from the Chrome Web Store
2

Open the extension

Click the MCP-B icon in your browser toolbar
3

Choose an agent

Select from Userscript Engineer, WebMCP Server, Browsing Agent, or Chat Companion
4

Start creating

Describe what you want to build and let the AI agent help you create it

​
Core Features

​
AI Agents

Four specialized agents help you accomplish different tasks:
Build custom scripts to enhance websitesPerfect for adding features, removing clutter, or automating repetitive tasks. The Userscript Engineer helps you create production-grade userscripts through conversation.

Learn About Agents

Understand when to use each agent

​
Userscript Management

Create, organize, and share custom scripts:

Create with AI

Describe what you want in plain language - the agents build it for you

Download & Upload

Export scripts to share with others or back up your library

Enable/Disable

Turn scripts on and off without deleting them

Edit & Update

Modify existing scripts through conversation or manual editing

Managing Userscripts

Complete guide to downloading, uploading, and organizing scripts

​
MCP Tool Integration

The extension implements the Model Context Protocol, enabling:
  • Tool Discovery: Automatically detect MCP tools on websites
  • Tool Execution: Call website tools from AI agents
  • Tool Creation: Build your own MCP servers as userscripts
  • Native MCP Support: Connect to traditional MCP servers via native host
WebMCP is a polyfill for the W3C Web Model Context API (navigator.modelContext) - a standards-track proposal for exposing website functionality to AI agents. The API supports Zod schemas (preferred) as well as JSON Schema for type-safe tool definitions.

​
How It Works

​
Architecture

​
Extension Components

Chat interface with specialized AI agents that understand web automation and MCP. Switch between agents based on your current task.
Built-in manager for creating, editing, enabling, and organizing userscripts. Export and import scripts to share with others.
Browser-specific implementation of the Model Context Protocol that works entirely in the browser without backend servers.
Suite of MCP tools for DOM inspection, tab management, userscript execution, and browser automation.
Optional component that connects browser MCP to traditional filesystem-based MCP servers.

​
Use Cases

​
Website Customization

Dark Modes

Add custom dark themes to any website

Remove Ads

Hide unwanted ads, banners, and distracting elements

Custom Layouts

Rearrange page elements to match your preferences

Keyboard Shortcuts

Add custom hotkeys for common actions

​
Automation

Form Filling

Auto-populate forms with saved information

Data Extraction

Pull structured data from websites automatically

Monitoring

Track price changes, content updates, or availability

Batch Actions

Process multiple items or pages in sequence

​
AI Integration

Search Tools

Let AI search websites and retrieve results

Action Tools

Enable AI to click buttons, submit forms, or navigate

Data Tools

Give AI access to structured website data

Workflow Tools

Chain multiple website actions into complex workflows

​
Example: Building a Dark Mode

Here’s how you’d use the extension to add dark mode to a website:
1

Select Userscript Engineer

Open the extension and choose “Userscript Engineer” from the agent dropdown
2

Navigate to the site

Go to the website you want to customize
3

Request the feature

Type: “Add a dark mode toggle button that changes the background to #1a1a1a and text to #e0e0e0”
4

Agent builds and tests

The agent inspects the page, writes TypeScript code, and tests it
5

Review and refine

See the changes live. Request adjustments if needed
6

Save the script

The agent commits the script. It will now run automatically on that site

​
Example: Creating MCP Tools

Here’s how to expose a website’s search as an MCP tool:
1

Select WebMCP Server

Open the extension and choose “WebMCP Server” from the agent dropdown
2

Navigate to the site

Go to the website with the functionality you want to expose
3

Request the tools

Type: “Create MCP tools for searching this site and returning results”
4

Agent analyzes and builds

The agent inspects the search form, writes tool registration code, and tests it
5

Tools become available

After testing, the tools appear in your MCP tool list
6

Use from any agent

Any AI agent can now search that website on your behalf

​
Browser Compatibility

✅ Fully supportedInstall from the Chrome Web Store

​
Privacy & Security

Userscripts run entirely in your browser. No data is sent to external servers except when you explicitly request AI agent interactions.
Always review userscripts before enabling them, especially from untrusted sources. Scripts have access to the websites they run on.
The extension only runs scripts on sites you specify. You control which websites each script can access.
Core MCP libraries are open source and available on GitHub for audit and contribution.
Userscripts can access and modify website content. Only install scripts from trusted sources and review code before importing.

​
Get Help

Understanding Agents

Learn which agent to use for each task

Managing Userscripts

Download, upload, and organize your scripts

Discord Community

Get help from other users and developers

GitHub Issues

Report bugs or request features

​
Architecture & Development

Dive deep into how the extension works internally, including background workers, content scripts, and the MCP transport layer.

View Extension Architecture

Technical architecture documentation
Build your own Chrome extensions using MCP wrappers for Chrome Extension APIs.

View API Documentation

@mcp-b/extension-tools package docs

​
Next Steps

1

Install the extension

Get it from the Chrome Web Store
2

Set up Native Host (Recommended)

Connect browser tools to desktop AI with the Native Host Setup guide - use WebMCP from Claude Code, Claude Desktop, and more!
3

Learn about agents

Read Understanding Agents to choose the right agent for your task
4

Create your first script

Try asking the Userscript Engineer to add a simple feature to your favorite website
5

Explore examples

Check out the Examples Repository for inspiration
6

Join the community

Share your creations on Discord