项目说明：此 MCP 服务器实现基于浏览器使用/Web UI基础。核心浏览器自动化逻辑和配置模式均改编自原始项目。
人工智能驱动的浏览器自动化服务器，实现模型上下文协议 (MCP)，用于自然语言浏览器控制。
🧠 MCP 集成- AI 代理通信的完整协议实现
🌐浏览器自动化- 页面导航、表单填写和元素交互
👁️视觉理解- 屏幕截图分析和基于视觉的交互
🔄状态持久化- 在任务之间维护浏览器会话
🔌多 LLM 支持- OpenAI、Anthropic、Azure、DeepSeek 集成
Related MCP server: Cloudflare Playwright MCP
Python 3.11 或更高版本
uv（快速 Python 包安装程序）
Chrome/Chromium浏览器
在 MacOS 上： ~/Library/Application\ Support/Claude/claude_desktop_config.json在 Windows 上： %APPDATA%/Claude/claude_desktop_config.json
"mcpServers": {
"browser-use": {
"command": "uvx",
"args": [
"mcp-server-browser-use",
],
"env": {
"OPENROUTER_API_KEY": "",
"OPENROUTER_ENDPOINT": "https://openrouter.ai/api/v1",
"OPENAI_ENDPOINT": "https://api.openai.com/v1",
"OPENAI_API_KEY": "",
"ANTHROPIC_ENDPOINT": "https://api.anthropic.com",
"ANTHROPIC_API_KEY": "",
"GOOGLE_API_KEY": "",
"AZURE_OPENAI_ENDPOINT": "",
"AZURE_OPENAI_API_KEY": "",
"DEEPSEEK_ENDPOINT": "https://api.deepseek.com",
"DEEPSEEK_API_KEY": "",
"MISTRAL_API_KEY": "",
"MISTRAL_ENDPOINT": "https://api.mistral.ai/v1",
"OLLAMA_ENDPOINT": "http://localhost:11434",
"ANONYMIZED_TELEMETRY": "true",
"BROWSER_USE_LOGGING_LEVEL": "info",
"CHROME_PATH": "",
"CHROME_USER_DATA": "",
"CHROME_DEBUGGING_PORT": "9222",
"CHROME_DEBUGGING_HOST": "localhost",
"CHROME_PERSISTENT_SESSION": "false",
"BROWSER_HEADLESS": "false",
"BROWSER_DISABLE_SECURITY": "false",
"BROWSER_WINDOW_WIDTH": "1280",
"BROWSER_WINDOW_HEIGHT": "720",
"BROWSER_TRACE_PATH": "trace.json",
"BROWSER_RECORDING_PATH": "recording.mp4",
"RESOLUTION": "1920x1080x24",
"RESOLUTION_WIDTH": "1920",
"RESOLUTION_HEIGHT": "1080",
"VNC_PASSWORD": "youvncpassword",
"MCP_MODEL_PROVIDER": "anthropic",
"MCP_MODEL_NAME": "claude-3-5-sonnet-20241022",
"MCP_TEMPERATURE": "0.3",
"MCP_MAX_STEPS": "30",
"MCP_USE_VISION": "true",
"MCP_MAX_ACTIONS_PER_STEP": "5",
"MCP_TOOL_CALL_IN_CONTENT": "true"
}
}"browser-use": {
"command": "uv",
"args": [
"--directory",
"/path/to/mcp-browser-use",
"run",
"mcp-server-browser-use"
],
"env": {
...
}
}# Install dev dependencies
uv sync
# Run with debugger
npx @modelcontextprotocol/inspector uv --directory . run mcp-server-browser-use浏览器冲突：启动前关闭所有 Chrome 实例。
API 错误：验证环境变量中的 API 密钥是否与您的 LLM 提供商匹配。
视觉支持：确保MCP_USE_VISION=true进行屏幕截图分析。
服务器通过环境变量支持多个 LLM 提供程序。以下是MCP_MODEL_PROVIDER的可用选项：
提供者 | 价值 | 必需的环境变量 |
人择 |
|
|
OpenAI |
|
|
Azure OpenAI |
|
|
DeepSeek |
|
|
双子座 |
|
|
米斯特拉尔 |
|
|
奥拉马 |
|
|
OpenRouter |
|
|
对于标记为可选的端点，如果未指定，则将使用默认值
可以使用MCP_TEMPERATURE配置温度（默认值：0.3）
可以使用MCP_MODEL_NAME指定模型
对于 Ollama 模型，可以配置其他上下文设置，例如num_ctx和num_predict
本项目在 MIT 许可证下扩展了browser-use/web-ui 。特别感谢原作者提供的浏览器自动化框架。
MIT——详情请参阅许可证。
Handle run-browser-agent tool calls.
| Name | Required | Description | Default |
|---|---|---|---|
| add_infos | No | ||
| task | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations provided, the description carries the full burden of behavioral disclosure but fails completely. It doesn't indicate whether this is a read or write operation, what side effects it might have, what permissions are required, or what the expected behavior is. The phrase 'Handle... tool calls' is too vague to convey any meaningful behavioral information.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
While the description is technically concise (only 5 words), this represents under-specification rather than effective brevity. The single sentence doesn't earn its place by providing meaningful information. A truly concise description would still convey essential purpose and usage information in minimal words, which this fails to do.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given a tool with 2 parameters, no annotations, no output schema, and 0% schema description coverage, the description is completely inadequate. It provides no information about what the tool does, how to use it, what parameters mean, or what to expect from its operation. The description fails to compensate for any of the missing structured information.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The schema description coverage is 0%, meaning neither parameter has any description in the schema. The tool description provides no information about what the 'task' or 'add_infos' parameters mean, what format they should take, or how they affect the tool's operation. For a tool with 2 parameters and zero schema documentation, this represents a complete failure to add parameter semantics.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description 'Handle run-browser-agent tool calls' is essentially a tautology that restates the tool name without explaining what the tool actually does. It doesn't specify what 'run-browser-agent' means, what resources it operates on, or what action it performs. While it mentions 'tool calls', this adds no meaningful information beyond the name itself.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description provides absolutely no guidance on when to use this tool, what context it's appropriate for, or what alternatives might exist. There are no sibling tools mentioned, but even for a standalone tool, the description fails to give any indication of its intended use case or prerequisites.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
v1.0.0run_browser_agentScored across 1 tool
With only one tool, there is no possibility of ambiguity or overlap between tools. The single tool 'run_browser_agent' has a distinct purpose that cannot be confused with any other tool in this set.
The naming pattern cannot be inconsistent with only one tool. The tool name 'run_browser_agent' follows a clear verb_noun pattern, and there are no other tools to create naming conflicts or mixed conventions.
A single tool is generally too few for most server purposes, as it limits functionality and may not cover the domain adequately. For a browser automation server, one tool feels thin and restrictive compared to typical expectations of multiple actions like navigation, clicking, or form filling.
The tool surface is severely incomplete for browser automation. With only one tool to run an agent, there are obvious gaps in core operations such as navigating to URLs, interacting with page elements, extracting data, or managing browser sessions, which will likely cause agent failures in practical use.
AI-powered browser automation — navigate, click, fill forms, and extract data from any website.
AI-powered web automation. Navigate websites using AI agents for one page or a thousand
AI-powered web automation. Navigate websites using AI agents for one page or a thousand
Stealth web automation for AI agents. Login, signup, navigate, screenshot.
