Daytona
@daytonaio
Run code in secure remote sandboxes
Overview
What is Daytona?
Daytona is an MCP server that enables AI agents to manage Daytona sandboxes, execute shell commands, perform file operations, clone Git repositories, and generate preview links for web applications running in sandboxes. It is intended for users of compatible AI agents such as Claude Desktop App, Claude Code, Cursor, and Windsurf who have a Daytona account and the Daytona CLI installed.
How to use Daytona?
Install the Daytona CLI (brew install daytonaio/cli/daytona on Mac/Linux, a PowerShell command on Windows), log in with daytona login, then run daytona mcp init [claude/cursor/windsurf] to configure the server for a specific agent. For other agents, run daytona mcp config and paste the resulting JSON into the agent’s MCP configuration.
Key features of Daytona
- Create and destroy Daytona sandboxes on demand
- Execute shell commands in ephemeral Linux sandboxes
- Upload, download, move, delete, and list files
- Clone Git repositories into sandboxes
- Generate secure preview URLs for web apps
Use cases of Daytona
- Provision and teardown sandboxes for isolated code execution
- Automate file editing and testing in AI agent workflows
- Run build, test, or deployment commands inside sandboxes
- Clone and work with remote repositories without local setup
- Expose web applications running in sandboxes via tunneling
FAQ from Daytona
What are the prerequisites to use the Daytona MCP server?
You need a Daytona account, the Daytona CLI installed, and a compatible AI agent (Claude Desktop App, Claude Code, Cursor, or Windsurf).
How do I authenticate the server?
Run daytona login in your terminal to refresh your Daytona credentials.
What should I do if I get connection errors?
Ensure that the Daytona MCP server is properly configured. Re-run daytona mcp init or verify the JSON configuration from daytona mcp config.
How can I check the status of my sandboxes?
Use the command daytona sandbox list to see all sandboxes and their states.
Does the server support AI agents other than those listed?
Yes. You can use daytona mcp config to get a JSON configuration that can be pasted into the MCP settings of any agent that supports the Model Context Protocol.