agentify-desktop

agentify-desktop

MCP server that enables AI tools to control local browser sessions for ChatGPT, Claude, and other AI services, supporting querying, navigation, file uploads, and artifact management.

Category
访问服务器

README

Agentify Desktop

Agentify Desktop is a local control center for AI web sessions. It lets MCP-capable tools such as Codex, Claude Code, and OpenCode use the AI subscriptions you are already signed into, while keeping browser state, files, and automation on your machine.

What It Does

  • Opens a local Agentify Control Center.
  • Manages signed-in browser sessions for ChatGPT, Claude, Perplexity, Gemini, Google AI Studio, and Grok.
  • Exposes MCP tools for querying a tab, reading a page, navigating, uploading files, saving artifacts, and reusing stable tab keys.
  • Supports parallel tabs so different agents or tasks can use separate sessions.
  • Packs local repo/file context into prompts when requested.
  • Saves generated images/files locally so they can be reused in follow-up prompts.

Example Prompts After MCP Setup

Once Agentify Desktop is running and registered with your MCP client, you can ask for workflows like:

  • “Use Agentify with key repo-triage to ask ChatGPT for a second opinion on this bug, then compare its answer with your own analysis.”
  • “Open a Perplexity tab with key research-auth-flow and research current OAuth best practices for desktop apps.”
  • “Send this implementation plan to Claude in a separate Agentify tab and summarize any risks it finds.”
  • “Use Agentify to generate three UI concept images, save the images as artifacts, and return the local file paths.”
  • “Open Grok and ChatGPT in separate Agentify tabs, ask both to review this API design, then compare the tradeoffs.”
  • “Pack this repo into context, ask ChatGPT to identify risky files, and save the conversation under a stable tab key for follow-ups.”
  • “Read the current ChatGPT page through Agentify and turn the conversation into actionable TODOs.”

Requirements

  • Node.js 20 or newer
  • An MCP-capable CLI if you want tool integration: Codex, Claude Code, or OpenCode

Supported Sites

  • chatgpt.com
  • claude.ai
  • perplexity.ai
  • aistudio.google.com
  • gemini.google.com
  • grok.com

Preferred Install And Run

Start the desktop GUI without cloning this repo:

npx @agentify/desktop

Equivalent explicit GUI command:

npx @agentify/desktop gui

If you prefer a global install:

npm install -g @agentify/desktop
agentify-desktop

If you want the older repo-clone and local source workflow, use DEVELOPMENT_FROM_SOURCE.md.

MCP Server

Run the MCP server over stdio:

npx @agentify/desktop mcp

Show newly-created browser tabs while debugging:

npx @agentify/desktop mcp --show-tabs

With a global install:

agentify-desktop-mcp
agentify-desktop-mcp --show-tabs

Register With MCP Clients

Codex:

codex mcp add agentify-desktop -- npx -y @agentify/desktop mcp

Claude Code:

claude mcp add --transport stdio agentify-desktop -- npx -y @agentify/desktop mcp

OpenCode config example:

{
  "mcp": {
    "agentify-desktop": {
      "type": "local",
      "command": ["npx", "-y", "@agentify/desktop", "mcp"],
      "enabled": true
    }
  }
}

Use --show-tabs at the end of the command while debugging:

codex mcp add agentify-desktop -- npx -y @agentify/desktop mcp --show-tabs

First Run

  1. Start the app:
npx @agentify/desktop
  1. In the Control Center, create or show a ChatGPT tab.

  2. Sign in to the target vendor in the browser window.

  3. Register the MCP server with your CLI.

  4. Ask your MCP client to use Agentify:

Use Agentify Desktop with tab key repo-triage.
Ask ChatGPT to summarize this repo in 8 bullets and list the top 3 risky areas to change first.
Return the answer and keep the tab key stable for follow-ups.

The core loop is:

  • keep a real signed-in browser session open locally
  • call it from an MCP client
  • reuse a stable tab key across follow-up prompts

Useful MCP Tools

The MCP server registers agentify_* tools, including:

  • agentify_query: send a prompt to a stable tab and return the assistant response.
  • agentify_read_page: read visible page text from a tab.
  • agentify_navigate: navigate a tab to a URL.
  • agentify_ensure_ready: wait for login, CAPTCHA, or UI readiness.
  • agentify_show / agentify_hide: bring windows forward or minimize them.
  • agentify_status: inspect tab and readiness state.
  • agentify_tabs, agentify_tab_create, agentify_tab_close: manage tabs.
  • agentify_save_artifacts, agentify_list_artifacts, agentify_open_artifacts_folder: manage generated files/images.
  • agentify_save_bundle, agentify_list_bundles: save and reuse context bundles.
  • agentify_add_watch_folder, agentify_list_watch_folders, agentify_remove_watch_folder: manage watched folders.

Artifact Workflow

Generate an image or file in a stable tab:

{
  "tool": "agentify_query",
  "arguments": {
    "key": "ui-concepts",
    "prompt": "Generate 3 clean UI concept images for a compact desktop developer tool. Keep backgrounds neutral and avoid text."
  }
}

Save the generated images locally:

{
  "tool": "agentify_save_artifacts",
  "arguments": {
    "key": "ui-concepts",
    "mode": "images",
    "maxImages": 3
  }
}

Reattach one of the returned file paths in a follow-up:

{
  "tool": "agentify_query",
  "arguments": {
    "key": "ui-concepts",
    "prompt": "Use the attached concept image and create a more minimal variant with stronger contrast.",
    "attachments": ["/absolute/path/to/concept.png"]
  }
}

Codebase Context Workflow

Ask Agentify to pack local files or folders into a prompt:

{
  "tool": "agentify_query",
  "arguments": {
    "key": "repo-review",
    "prompt": "Summarize this codebase in 8 bullets and list the top 3 risky files to change first.",
    "contextPaths": ["/absolute/path/to/repo"]
  }
}

Control context size:

{
  "tool": "agentify_query",
  "arguments": {
    "key": "repo-review",
    "prompt": "Focus only on rendering and state management.",
    "contextPaths": ["/absolute/path/to/repo"],
    "maxContextChars": 120000,
    "maxContextFiles": 80,
    "maxContextInlineFiles": 30
  }
}

The tool result includes packedContextSummary so you can see what was included, attached, or skipped.

Browser Backend

Agentify Desktop supports two browser backends:

  • chrome-cdp: launches or attaches to a Chrome-family browser over Chrome DevTools Protocol. This is the default and recommended backend.
  • electron: embedded windows managed by Agentify Desktop. Use this only as an explicit fallback.

Chrome CDP is the default because SSO providers commonly block embedded Electron login:

npx @agentify/desktop

Optional Chrome CDP settings:

AGENTIFY_DESKTOP_CHROME_DEBUG_PORT=9333 npx @agentify/desktop
AGENTIFY_DESKTOP_CHROME_BIN="/Applications/Google Chrome.app/Contents/MacOS/Google Chrome" npx @agentify/desktop

You can also pass GUI flags:

npx @agentify/desktop gui --browser-backend chrome-cdp
npx @agentify/desktop gui --browser-backend electron
npx @agentify/desktop gui --chrome-debug-port 9333

Chrome CDP profile modes:

  • Agentify isolated profile: safest default.
  • Existing Chrome profile: reuses your normal Chrome session. Fully quit Chrome first so the profile is not already locked.

CAPTCHA And Login Policy

Agentify Desktop does not bypass CAPTCHAs or use third-party solvers. If a verification or login challenge appears, automation pauses, brings the relevant window forward, and waits for you to complete the step manually.

If your account uses Google, Microsoft, or Apple SSO, keep auth popups enabled in the Control Center. If embedded login remains unreliable, use Chrome CDP.

Windows Notes

Use Node.js 20 or 22 on Windows. Agentify Desktop is tested against Windows in CI, including the npm CLI launcher path.

Chrome CDP is still the recommended backend on Windows because Google and Microsoft SSO can block embedded Electron login. Agentify looks for Chrome, Chromium, Brave, and Microsoft Edge in the usual install locations and on PATH.

If Chrome CDP cannot find your browser, set the executable explicitly:

$env:AGENTIFY_DESKTOP_CHROME_BIN = "C:\Program Files\Google\Chrome\Application\chrome.exe"
npx @agentify/desktop

Local Data And Privacy

Agentify Desktop is local-first:

  • The local API binds to 127.0.0.1.
  • The local API requires a bearer token stored under ~/.agentify-desktop/.
  • Electron browser data is stored under ~/.agentify-desktop/electron-user-data/.
  • Chrome CDP profile data is stored under ~/.agentify-desktop/chrome-user-data/ unless you choose an existing profile.
  • Artifacts, bundles, logs, and state are stored under ~/.agentify-desktop/.

Anyone with access to your machine account may be able to access local session data. Treat the machine account as the security boundary.

Environment Variables

  • AGENTIFY_DESKTOP_STATE_DIR: override the local state directory.
  • AGENTIFY_DESKTOP_PORT: choose the local API port.
  • AGENTIFY_DESKTOP_SHOW_TABS=true: show newly-created tabs by default.
  • AGENTIFY_DESKTOP_MAX_TABS: cap parallel tabs.
  • AGENTIFY_DESKTOP_BROWSER_BACKEND=electron|chrome-cdp: choose browser backend.
  • AGENTIFY_DESKTOP_CHROME_BIN: choose Chrome/Chromium executable.
  • AGENTIFY_DESKTOP_CHROME_DEBUG_PORT: choose Chrome debug port.
  • AGENTIFY_DESKTOP_CHROME_PROFILE_MODE=isolated|existing: choose Chrome profile mode.
  • AGENTIFY_DESKTOP_CHROME_PROFILE_NAME: choose an existing Chrome profile name.

Development From Source

Source checkout, quickstart script usage, local build commands, and source-only debugging notes live in DEVELOPMENT_FROM_SOURCE.md.

Package Commands

The npm package exposes these commands:

  • agentify-desktop: default GUI launcher, with mcp subcommand support.
  • agentify-desktop-gui: explicit GUI alias.
  • agentify-desktop-mcp: explicit MCP alias.

Examples:

npx @agentify/desktop
npx @agentify/desktop mcp
npx -p @agentify/desktop agentify-desktop-mcp

License And Trademarks

The code is licensed under MPL-2.0. Agentify trademarks and branding are not included in that license. See TRADEMARKS.md.

推荐服务器

Baidu Map

Baidu Map

百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。

官方
精选
JavaScript
Playwright MCP Server

Playwright MCP Server

一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。

官方
精选
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。

官方
精选
本地
TypeScript
VeyraX

VeyraX

一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。

官方
精选
本地
graphlit-mcp-server

graphlit-mcp-server

模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。

官方
精选
TypeScript
Kagi MCP Server

Kagi MCP Server

一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

Exa MCP Server

模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。

官方
精选