跳到主要内容
Supermarket
返回能力市场
MCP Server
programming
Apache-2.0

mcp

Browser MCP is a Model Context Provider (MCP) server that allows AI applications to control your browser

BrowserMCPBrowserMCP
50/ 100

公开评测 · 综合采用结论

具备基础能力,但文档或工程质量仍需完善

查看评测依据 查找我项目的报告基于公开项目证据,非安全认证或安装推荐
7.2kstars
592forks
最近更新 21天前
评测生成时间(北京时间)
本报告引擎
v3.10.0
当前引擎
v3.17.3

本报告与当前引擎使用不同规则;原分数不会自动更新,不同版本的分数不宜直接对比。

重新评测此项目

进入后确认来源与额度,提交才会创建任务。

代码版本未记录或记录无效,无法核对本报告的 commit 与送评文本;不会用当前项目代码回填历史记录。

Evaluation report

综合采用结论

50
D
满分 100 · 静态采用参考
需要完善低风险
决策摘要

具备基础能力,但文档或工程质量仍需完善

66%
中置信度
25
文档
100
安全
38
质量
12
活跃
67
采用
  • 基础评测完成+25/25确定性评分与静态安全扫描已完成
  • README 有效证据+3/251,198 个去重后的有效字符
  • 独立证据来源+8/202 类非重复证据,重复文件不叠加
  • 仓库元数据+10/10已取得仓库状态与采用数据
  • 活跃记录+5/5已取得最近提交时间
  • AI 复核+15/15已完成结构化 AI 证据复核

本报告评了什么

综合分用于静态采用筛选,不是任务成功率,也不代表使用 Skill 后的能力提升。

静态检查范围
记录 2 个文件 · 2,664 字符
只检查采集到的受限文本与已知风险模式;未读取的脚本、依赖和运行环境不在结论内。
AI 内容复核
已使用
复核送评文本的说明和设计,不是安装测试或实际任务执行。记录缺失时不推断已使用或未使用。
真实任务效果
本报告未提供实测证据
没有使用/不使用 Skill 的任务对照、重复运行或耗时与 token 成本结果。采用前仍需在隔离环境验证自己的任务。

脚本引用覆盖

脚本引用覆盖记录不可得;不推断没有引用或已经检查。

How it works · 未生成图示

公开证据不足,暂不绘图

当前材料不足以同时核实至少 2 个步骤或组件及 1 条关系。为避免臆测,报告保留文字证据,不生成流程图、时序图或架构图。

补充参与方或组件说明谁参与、各自负责什么
说明关系与顺序提供输入输出、调用或依赖证据
重新评测自动选图按证据选择流程、时序或架构图
五维表现
解决浏览器自动化真实问题,但缺少安装、配置、示例和限制说明,无法直接采用。
质量证据
  • README中'About'描述为MCP server + Chrome extension
  • Features列出本地执行、隐私、登录状态、隐身特性
  • Contributing说明无法独立构建,依赖monorepo
  • Credits提及改编自Playwright MCP
采用建议
优势
  • 问题与用途描述
  • 有效 README
  • 结构化章节
  • 未发现已知高风险模式
  • 明确解决浏览器自动化问题
关注点
  • 缺少安装或接入步骤
  • 缺少可执行示例
  • 缺少输入、参数或工具说明
  • 无安装或接入步骤
  • 无可执行示例
适合

需要浏览器自动化的AI应用用户、重视隐私和本地执行的场景、希望利用现有登录状态的用户、寻求避免基本bot检测的用户

不建议直接用于

需要详细配置和示例的快速集成、需要明确错误处理和排障的场景

也有自己的公开项目?先免登录查已有报告;没有报告时,可保留地址继续登录评测。

查找我项目的报告 →
文档证据
25/100
问题与用途描述10 分
有效 README12 分
安装或接入步骤14 分
可执行示例16 分
输入、参数或工具说明11 分
输出或结果说明9 分
限制、权限或边界12 分
错误处理或排障8 分
许可证信息5 分
结构化章节3 分
安全证据
低风险
未发现已知高风险模式

静态扫描不是安全保证,生产接入前仍应人工复核权限和数据边界。

优先改进清单
  1. 01补充安装或接入步骤
  2. 02补充可执行示例
  3. 03补充输入、参数或工具说明
方法、证据与局限展开
数据来源

GitHub Repository API

扫描范围

2 个文件 · 2,664 字符

评测引擎

v3.10.0 · AI 内容复核:已使用(deepseek-chat)

局限
  • 静态评测不会安装或执行项目代码
  • 安全扫描基于高信号文件与已知模式,不能替代人工审计
  • 流行度只反映采用程度,不代表安全或工程质量

30 天热度趋势

README

Browser MCP

Browser MCP lets AI apps (Claude Code, Cursor, VS Code, Codex, Claude Desktop…) use your own Chrome: your signed-in sites, your profile, in tabs of their own. Everything runs on your computer.

How it works

Each AI app runs the MCP server (npx @browsermcp/mcp@latest), a small process that starts one daemon per user and forwards the app's browser tool calls to it. The daemon drives the browser's tabs with Playwright, through the extension: a thin chrome.debugger relay that only talks to the daemon on 127.0.0.1, and only after you click Allow in its popup once per computer, having checked that its six-digit code matches the one your AI app shows. The extension enforces its own rules whatever the daemon asks: no incognito tabs unless you turned on Allow in Incognito for the extension in Chrome, no browser or extension pages, and nothing while you've pressed Stop.

FolderPackage
packages/mcp@browsermcp/mcpThe MCP server and the daemon (README)
packages/protocol@browsermcp/protocolThe protocol between the extension and the daemon, including pairing
packages/extension-core@browsermcp/extension-coreThe extension's browser control: the connection, pairing, the debugger relay, the rules about tabs, tab groups, the badge
apps/extension@browsermcp/extensionThe extension built from this repo: browser control and its popup

Build it yourself

You need Node.js 24 or later, pnpm (corepack enable) and Chrome or another Chromium browser (Edge, Brave, Arc…). Firefox has no debugger API, so Browser MCP doesn't work there.

The Chrome Web Store version updates itself. A build you load yourself doesn't: you update it by hand (see Updating).

  1. Build the extension and the server:

    git clone https://github.com/BrowserMCP/mcp.git browsermcp
    cd browsermcp
    pnpm install
    pnpm build
    
  2. Open chrome://extensions, turn on Developer mode, click Load unpacked and pick apps/extension/.output/chrome-mv3. It's called "Browser MCP (self-built)". If the Chrome Web Store's Browser MCP is installed in the same profile, turn it off, or both would connect to the server as two browsers. Other Chrome profiles with Browser MCP connect too: agents can use tabs in each.

  3. In your AI app's MCP config, run the server you built instead of the one on npm:

    {
      "mcpServers": {
        "browsermcp": {
          "command": "node",
          "args": ["/path/to/browsermcp/packages/mcp/dist/index.js"]
        }
      }
    }
    

    npx @browsermcp/mcp@latest (the docs' setup) works with this extension too: it's the same server, published from this repo.

  4. Restart the AI app and ask it to do something in the browser. The first time, the extension's popup asks you to Allow the server on this computer: check that its code matches the one the AI app shows.

Every build of this extension has the same ID (imlplcjlhgpajopdbpbngiadkcmdbgob, from the key in apps/extension/extension-key.ts), which the server accepts besides the Chrome Web Store's. A build with another key needs its ID in BROWSER_MCP_EXTENSION_IDS (comma-separated) in the server's environment, or on a line of ~/.browsermcp/dev-extension-ids.

Updating

Only for a self-built extension: the Chrome Web Store version updates itself.

git pull
pnpm install
pnpm build

Then click the reload button on "Browser MCP (self-built)" in chrome://extensions, and restart your AI apps if they run the server from this folder. The extension and the server tell you when the other side is too old for them.

Development

  • pnpm lint, pnpm format (pnpm format:fix fixes it): Oxlint and Oxfmt.
  • pnpm typecheck, pnpm test: TypeScript and the unit tests (Vitest) of every package. The server's tests run Playwright's Chromium: install it once with pnpm --filter @browsermcp/mcp exec playwright-core install chromium.
  • pnpm smoke: builds the extension in development mode and the server, loads the extension in headless Chromium and drives it through the server as an AI app would, on random ports and a temporary BROWSER_MCP_HOME, so it never touches your own Browser MCP. Screenshots and the daemons' logs go to apps/extension/.output/smoke/ (--out <dir> elsewhere, e.g. for runs at once); a failed section names the step or check that failed, its error and the last call's output, and screenshots the open pages.
  • pnpm --filter @browsermcp/extension build:dev: a development build (apps/extension/.output/chrome-mv3-dev).

Contributing and security

See CONTRIBUTING.md for how to report bugs and send changes, and SECURITY.md to report a vulnerability privately.

License

Apache-2.0. Building it for yourself or your team is fine, and so is saying that your tool is compatible with Browser MCP or built on its code. If you publish a modified extension or server (to a browser extension store, npm or anywhere else others install it from), please give it another name and icon, and don't present it as the official Browser MCP or as endorsed by it, so people can tell the two apart. Keep the LICENSE and say what you changed.