AgentDock
AgentDock connects MCP-compatible AI clients to your computers and servers so they can work with files, run commands, use Git and browsers, and connect other services.
AgentDock itself is not a chat app or AI model. The connected AI client decides what to do; AgentDock performs the allowed work on the machine where it runs. Access is limited by the operating-system user, Docker mounts, and the capabilities you enable.
First use
Choose the installation guide for your current environment:
| Current environment | Recommended guide |
|---|---|
| macOS | macOS installation |
| Windows 11 | Windows installation |
| Linux server or desktop | Linux installation |
| Docker already installed | Docker installation |
After installation, add the MCP URL and token shown in the guide to your client using Streamable HTTP.
What can it do?
Describe the goal directly, for example:
- “Find out why this project fails to start, fix it, and verify the result.”
- “Read this repository, change the code, and commit it.”
- “Open a website, complete a lookup, and save the result as a file.”
- “Install and use a Skill, review a Plugin, or connect an external MCP service.”
- “Use the browser, desktop automation, or a coding tool such as Codex, Claude, or Grok.”
- “Break this long task into steps and keep its progress updated.”
The agent selects the appropriate tools based on your request.
If you want web ChatGPT to operate your computer, follow Connect with ChatGPT after installing with a public address.
Where data is stored
A native installation uses these defaults:
~/.agentdock AgentDock state, tasks, Skills, and configuration
~/AgentDock Default working directory
Docker stores both categories in Docker volumes by default. Removing the program or container does not necessarily remove its data; confirm what must be preserved before cleanup.