M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
O

Open Cowork

Community
Not a standard server
Open-source personal AI agent desktop app with one-click install, multi-model support, sandbox isolation, and MCP integration.

This is not a typical connect-and-use MCP server; read the source repository before using it.

How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #147 of 438
Stars
★ 2.2k Very popular
Transport
stdio (local process)
Runtime
Node.js 18+ · Prebuilt binary
Credentials
API key / credential required
License
MIT
Last commit
Tools
5
58FMRS · C

Open Cowork is a feature-rich AI agent desktop app suitable for intermediate users who want to automate office tasks with AI in a local environment. It offers strong security features (sandbox isolation), flexible model choices, and multi-platform support. However, it may not be suitable for Linux users, and requires users to have some configuration knowledge to set up API keys and sandboxing. Overall, it is a valuable tool, especially for those needing document generation and desktop automation.

Strongest · Maintenance 14/20 Weakest · Reliability 8/20

Reliability
8/20
Security and permissions
12/20
Maintenance
14/20
Documentation
13/20
Setup experience
11/20
Why each score
Reliability 8/20
Evidence shows an active repository with a clear architecture, but static review found no test files or CI configuration, so actual startup and tool-list handshake cannot be verified. The README describes features, but there is no evidence that MCP initialization paths are tested. Deductions: lack of test evidence and error handling details.
Security and permissions 12/20
The repository uses an MIT license, and the README clearly discloses sandbox isolation mechanisms (WSL2, Lima) and path restrictions. However, no specific implementation of credential management was found, and remote control features (Feishu/Slack) and GUI operations may increase the attack surface. Deductions: lack of evidence for least privilege principles, and dangerous operations (e.g., file deletion) do not explicitly require confirmation. Red lines not met.
Maintenance 14/20
The repository has a clear MIT license, and star count and open issues indicate community activity. However, static review did not find release records, contribution guidelines, or a security response policy. Deductions: lack of evidence for detailed maintenance processes and version control.
Documentation 13/20
The README provides installation, configuration, feature overview, and architecture diagrams, but lacks detailed tool parameters, API reference, and troubleshooting guides. Deductions: documentation is incomplete, does not disclose model limitations, cost information, or error handling processes.
Setup experience 11/20
Multiple platform installation options are provided (Homebrew, installers, source build), but static review could not verify the reliability of installation scripts. Setup requires multiple manual steps (getting API key, configuring Base URL), and client configuration examples are insufficient. Deductions: installation steps are complex, and detailed examples for client connection are lacking.

Static review · not runListed 2026-08-07

Read the FMRS scoring method →

Fit and risk

What it can accessReads local filesWrites / deletes local filesRuns commands or codeUses the networkControls a browser

Best for

  • Users needing AI-powered desktop automation on Windows or macOS.
  • Users who want a GUI to use AI coding tools like Claude Code.
  • Security-conscious users who need sandbox isolation.
  • Users who need to generate professional document formats (PPTX, DOCX).
  • Teams that want to integrate remote control via Feishu or Slack.

Not for

  • Linux users, as no official installer is provided (though you can build from source).
  • Advanced users who prefer command-line tools without a GUI.
  • Users with no API keys or who are sensitive to API costs.
  • Users needing fully offline operation, as AI functionality requires external API calls.
  • Users wanting to use non-standard models or custom endpoints, though OpenAI-compatible APIs are supported, extra configuration may be needed.

Required permissions

  • Read and write access to the selected workspace folder.
  • Ability to execute commands in VMs (WSL2 or Lima).
  • Access to external services via MCP connectors (e.g., browser, Notion).
  • Network access to call AI model APIs and external services.

Risks and side effects

  • AI operations may accidentally delete or modify files; use sandbox isolation and authorize carefully.
  • If sandbox is disabled, commands execute locally on the host, posing security risks.
  • API keys are stored locally; protect them from exposure.
  • Remote control features (like Feishu) can be abused; ensure access controls.
  • Third-party MCP connectors may introduce security risks; only use trusted ones.

Setup

Before you start

Runtime：Node.js 18+ · Prebuilt binary

  1. Download the installer for your platform from the Releases page (Windows .exe or macOS .dmg).
  2. Run the installer and follow the prompts.
  3. Open the app and configure your AI model API key in settings (e.g., OpenRouter, Anthropic, or Chinese models).
  4. Select a workspace folder and start using.
  5. For macOS, you can also use Homebrew: brew tap OpenCoworkAI/tap && brew install --cask --no-quarantine open-cowork.

Check that it works

After installing the Open Cowork desktop app, paste an API key and Base URL in Settings, then ask it to generate a PPTX; if the Trace Panel shows the pptx skill executing and the file appears in your workspace, the setup works.

Troubleshooting

  1. If you encounter a security warning on macOS, go to System Settings > Privacy & Security and click 'Open Anyway', or install via Homebrew to bypass.
  2. If web tools (like WebSearch) fail, check proxy settings and enable TUN mode if needed.
  3. For Notion connector, besides setting the integration token, you must add connections in a root page.
  4. Ensure API key and Base URL are set correctly, especially for Chinese models (GLM, MiniMax).
  5. If sandbox VM is not active, check whether WSL2 (Windows) or Lima (macOS) is installed.

Things to try

Once connected, you can ask your AI assistant things like:

  • Read the financial_report.csv in this folder and create a PowerPoint summary with 5 slides.
  • Organize the meeting notes in my workspace into a Word document.
  • Turn this data into an Excel spreadsheet.
  • Use skill-creator to build a custom skill for processing invoices.

Tools 5

pptx writes
Built-in skill for generating and editing PowerPoint presentations.
docx writes
Built-in skill for processing and generating Word documents.
pdf writes
Built-in skill for handling and creating PDF files and forms.
xlsx writes
Built-in skill for handling Excel spreadsheets.
skill-creator writes
Built-in tool for creating custom skills.

Use cases

Create and edit presentations, documents, spreadsheets, and PDF files.
Automate file management tasks such as folder organization.
Control desktop applications through computer use.
Remote control of the AI agent via Feishu or Slack.
Extend AI capabilities using MCP connectors like browser and Notion.

Supported clients

Claude Desktop

Listed from the project's documentation, not tested by this site.

Overview

Open Cowork is a free, open-source AI agent desktop application for Windows and macOS. It wraps Claude Code, OpenAI, Gemini, DeepSeek, and other AI models into a user-friendly GUI with one-click installation — no coding required. Key capabilities include VM-level sandbox isolation (WSL2 on Windows, Lima on macOS), a built-in Skills system for generating PPTX, DOCX, XLSX, and PDF documents, MCP (Model Context Protocol) integration for connecting to browsers, Notion, and other desktop apps, GUI automation via computer use, and remote control through Feishu (Lark) and Slack. Open Cowork is the open-source implementation of Claude Cowork, designed to make AI-powered desktop automation accessible to everyone. It is built with Electron and the source is hosted at the repository root.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 62.9k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision 910467306e95 Data synced 2026-10-11 Read the FMRS scoring method