The Browser Agent Protocol (BAP) is an open standard designed for AI agents to interact with web browsers. It provides two interfaces, CLI and MCP, and supports features such as semantic selectors, composite actions, and structured data extraction.
2.5 points
4.6K

What is the BAP MCP Server?

The BAP MCP server is a tool based on the Model Context Protocol that allows AI assistants (such as Claude, Codex, etc.) to directly control web browsers. It provides 23 browser operation tools, including navigating web pages, clicking elements, filling out forms, taking screenshots, extracting data, etc., enabling AI assistants to browse and operate web pages like humans.

How to use the BAP MCP Server?

After installing the BAP MCP server, the AI assistant will automatically gain browser control capabilities. You can tell the assistant what to do through natural language, such as 'Open Hacker News and summarize the first 5 news stories', and the assistant will use the BAP tools to automatically complete these operations.

Applicable Scenarios

BAP is suitable for scenarios that require automated web page operations: data collection, web page testing, automated tasks, content analysis, price monitoring, news aggregation, form filling, web page screenshotting, etc.

Main Features

Composite Actions
Combine multiple steps (such as navigation + observation + click) into a single command for execution, saving 40 times the tokens compared to step-by-step operations.
Fused Operations
Combine navigation + observation, operation + observation, etc. into a single server call, reducing round-trip communication by 50 - 85%.
Semantic Selectors
Locate elements by their purpose (such as 'role:button:"Submit"') rather than their position, ensuring normal operation even if the web page is redesigned.
Structured Extraction
Extract structured data from any web page using JSON Schema to ensure the correct data format.
Accessibility-First Design
Built based on the accessibility tree inspection, designed specifically for AI understanding, providing clear page structure information.
Element References
Provide stable element references (such as @e1, e15) to maintain consistency between multiple observations.
Screenshot Annotation
Provide a numbered marker overlay on the screenshot for visual models, facilitating AI identification of page elements.
Advantages
No API key is required, and it runs completely locally to protect privacy.
Supports multiple browsers: Chrome, Firefox, WebKit, Edge.
Provides 23 tools to cover common browser operation needs.
Semantic selectors make operations more stable and less affected by page layout changes.
Composite actions significantly reduce the thinking steps and token consumption of AI assistants.
Supports mainstream AI platforms: Claude, Codex, Gemini, Cursor, etc.
Limitations
Requires local installation of Node.js and npm/npx.
The browser needs to be downloaded for the first run (about 100 - 300MB).
Complex web pages may require manual adjustment of selectors.
JavaScript-intensive pages may require additional waiting time.
The AI assistant needs to support the MCP protocol to use all functions.

How to Use

Install the BAP MCP Server
Install globally via npm or run directly using npx.
Configure the AI Assistant
Add the BAP MCP server to the configuration according to the AI assistant platform you are using.
Start Using
Restart the AI assistant. Now you can control the browser through natural language.

Usage Examples

News Aggregation and Summarization
Automatically access news websites, collect top news and generate summaries.
Automated Form Filling
Automatically log in to websites or fill out online forms.
Price Monitoring and Comparison
Extract product price information from e-commerce websites.
Web Page Screenshotting and Archiving
Automatically take screenshots of web page content for archiving or sharing.

Frequently Asked Questions

Does BAP require an internet connection?
Which browsers are supported?
How to select a specific element on the page?
Can BAP handle websites that require login?
How much disk space is required for installation?
Is BAP secure? Will it save my browsing data?

Related Resources

GitHub Repository
Source code, issue tracking, and contribution guidelines for BAP.
npm Package Page
npm package information and version history for the MCP server.
CLI Documentation
Detailed usage instructions and examples for the command-line interface.
MCP Protocol Documentation
Tool reference and configuration instructions for the MCP server.
Python SDK
Python version of the BAP client SDK.

Installation

Copy the following command to your Client for configuration
{
  "mcpServers": {
    "bap-browser": {
      "command": "npx",
      "args": ["-y", "@browseragentprotocol/mcp"]
    }
  }
}
Note: Your key is sensitive information, do not share it with anyone.

Alternatives

V
Vestige
Vestige is an AI memory engine based on cognitive science. By implementing 29 neuroscience modules such as prediction error gating, FSRS - 6 spaced repetition, and memory dreaming, it provides long - term memory capabilities for AI. It includes a 3D visualization dashboard and 21 MCP tools, runs completely locally, and does not require the cloud.
Rust
4.8K
4.5 points
M
Moltbrain
MoltBrain is a long-term memory layer plugin designed for OpenClaw, MoltBook, and Claude Code, capable of automatically learning and recalling project context, providing intelligent search, observation recording, analysis statistics, and persistent storage functions.
TypeScript
5.4K
4.5 points
B
Bm.md
A feature-rich Markdown typesetting tool that supports multiple style themes and platform adaptation, providing real-time editing preview, image export, and API integration capabilities
TypeScript
4.9K
5 points
S
Security Detections MCP
Security Detections MCP is a server based on the Model Context Protocol that allows LLMs to query a unified security detection rule database covering Sigma, Splunk ESCU, Elastic, and KQL formats. The latest version 3.0 is upgraded to an autonomous detection engineering platform that can automatically extract TTPs from threat intelligence, analyze coverage gaps, generate SIEM-native format detection rules, run tests, and verify. The project includes over 71 tools, 11 pre-built workflow prompts, and a knowledge graph system, supporting multiple SIEM platforms.
TypeScript
6.2K
4 points
P
Paperbanana
Python
7.4K
5 points
B
Better Icons
An MCP server and CLI tool that provides search and retrieval of over 200,000 icons, supports more than 150 icon libraries, and helps AI assistants and developers quickly obtain and use icons.
TypeScript
6.0K
4.5 points
A
Assistant Ui
assistant - ui is an open - source TypeScript/React library for quickly building production - grade AI chat interfaces, providing composable UI components, streaming responses, accessibility, etc., and supporting multiple AI backends and models.
TypeScript
6.7K
5 points
A
Apify MCP Server
The Apify MCP Server is a tool based on the Model Context Protocol (MCP) that allows AI assistants to extract data from websites such as social media, search engines, and e-commerce through thousands of ready-to-use crawlers, scrapers, and automation tools (Apify Actors). It supports OAuth and Skyfire proxy payment and can be integrated into MCP clients such as Claude and VS Code through HTTPS endpoints or local stdio.
TypeScript
7.7K
5 points
N
Notion Api MCP
Certified
A Python-based MCP Server that provides advanced to-do list management and content organization functions through the Notion API, enabling seamless integration between AI models and Notion.
Python
21.5K
4.5 points
M
Markdownify MCP
Markdownify is a multi-functional file conversion service that supports converting multiple formats such as PDFs, images, audio, and web page content into Markdown format.
TypeScript
34.5K
5 points
G
Gitlab MCP Server
Certified
The GitLab MCP server is a project based on the Model Context Protocol that provides a comprehensive toolset for interacting with GitLab accounts, including code review, merge request management, CI/CD configuration, and other functions.
TypeScript
24.7K
4.3 points
D
Duckduckgo MCP Server
Certified
The DuckDuckGo Search MCP Server provides web search and content scraping services for LLMs such as Claude.
Python
72.5K
4.3 points
F
Figma Context MCP
Framelink Figma MCP Server is a server that provides access to Figma design data for AI programming tools (such as Cursor). By simplifying the Figma API response, it helps AI more accurately achieve one - click conversion from design to code.
TypeScript
64.6K
4.5 points
U
Unity
Certified
UnityMCP is a Unity editor plugin that implements the Model Context Protocol (MCP), providing seamless integration between Unity and AI assistants, including real - time state monitoring, remote command execution, and log functions.
C#
32.3K
5 points
M
Minimax MCP Server
The MiniMax Model Context Protocol (MCP) is an official server that supports interaction with powerful text-to-speech, video/image generation APIs, and is suitable for various client tools such as Claude Desktop and Cursor.
Python
49.3K
4.8 points
G
Gmail MCP Server
A Gmail automatic authentication MCP server designed for Claude Desktop, supporting Gmail management through natural language interaction, including complete functions such as sending emails, label management, and batch operations.
TypeScript
21.1K
4.5 points
AIBase
Zhiqi Future, Your AI Solution Think Tank
© 2026AIBase