DeployKar MCP server

Deploy, monitor and debug your DeployKar apps from your IDE's AI chat: Kiro, Cursor, VS Code (Copilot), Claude Code, Windsurf, or any other MCP client.

"Deploy my app" · "Why did my build fail?" · "Set DATABASE_URL to postgres://…" · "Scale the API to 3 instances"

Your IDE's assistant does the thinking. This server only translates its tool calls into DeployKar API requests, using a personal API token.

Setup (2 minutes)

You need:

• uv installed (it provides the uvx command)
• a DeployKar account. Sign in at http://187.127.179.204:7100
• your app's code in a GitHub repository that DeployKar can access (connect GitHub once in the dashboard)

1. Sign in

First sign in to the DeployKar dashboard in your browser, then run:

uvx deploykar-mcp login

Your browser opens DeployKar's approval page. Choose Deploy (recommended) or Read-only access and click Approve. The token is saved on your computer, readable only by you: ~/.config/deploykar/credentials.json (Windows: %APPDATA%\deploykar\credentials.json).

2. Add DeployKar to your IDE

For Claude Code, create a file named .mcp.json in your project folder (or run claude mcp add --scope user deploykar -- uvx deploykar-mcp@latest to add it for all projects):

{
  "mcpServers": {
    "deploykar": {
      "command": "uvx",
      "args": ["deploykar-mcp@latest"]
    }
  }
}

The same JSON works in other IDEs:

VS Code (Copilot agent mode) uses .vscode/mcp.json with a slightly different shape:

{
  "servers": {
    "deploykar": {
      "type": "stdio",
      "command": "uvx",
      "args": ["deploykar-mcp@latest"]
    }
  }
}

Open the folder in your IDE, approve the deploykar server when asked, and start chatting. @latest means you get new versions automatically when your IDE restarts the server.

Without the login command (CI, remote machines, or if you prefer to paste a token): create a token on DeployKar's API tokens page (https://dev-32786578325.deploykar.com/auth/api/v1/api-tokens/authorize) and add it to the config. Don't commit a file that contains a token.

{
  "mcpServers": {
    "deploykar": {
      "command": "uvx",
      "args": ["deploykar-mcp@latest"],
      "env": { "DEPLOYKAR_API_TOKEN": "dk_live_..." }
    }
  }
}

What you can ask

• "Deploy this repo to DeployKar as a frontend"
• "Deploy my latest changes" (the assistant pushes first, then deploys and waits for the live URL)
• "Turn on auto-deploy for main"
• "Why did my last deploy fail? Fix it."
• "Show me the API's logs from the last few minutes"
• "Set DATABASE_URL to … and restart"
• "Scale the API to 3 instances"

Tools

How deployments work

DeployKar builds from GitHub, not from your local files, exactly like deploying from the dashboard. Whatever is pushed to the project's branch is what gets deployed, so the assistant commits and pushes your changes (after asking you) before deploying. With auto-deploy enabled, every push to the branch deploys automatically and the assistant just follows the deployment for the commit it pushed.

Security

• API tokens are stored hashed by DeployKar, shown once, expire after 30, 90 or 365 days, and can be revoked at any time: uvx deploykar-mcp logout, or the API tokens page (https://dev-32786578325.deploykar.com/auth/api/v1/api-tokens/manage).
• Tokens can only reach a fixed set of endpoints. They can never delete projects or deployments, remove backends, or manage databases, domains, secret groups or other tokens.
• Read-only tokens can view projects, statuses and logs, but cannot deploy or read environment variables.
• Environment variable values are never returned to the assistant.
• Per-token rate limits apply (120 requests/minute, 30 deploy or scale actions/hour).

Commands

uvx deploykar-mcp            run the MCP server (your IDE does this)
uvx deploykar-mcp login      sign in via the browser (--no-browser to paste a token instead)
uvx deploykar-mcp whoami     show the current account and token
uvx deploykar-mcp logout     revoke and delete the saved token

Environment variables (all optional):

Troubleshooting

• "Not signed in" / "rejected the API token": run uvx deploykar-mcp login again, then restart the MCP server in your IDE.
• The approval page asks you to sign in: sign in at http://187.127.179.204:7100 in the same browser, then reload the approval page.
• The server shows "Failed" or "Connection closed", or the IDE says uvx was not found: your IDE can't find uvx. Restart the IDE completely after installing uv. If that doesn't help, put the full path to uvx in "command":
  • macOS/Linux: run which uvx and use what it prints, e.g. /Users/you/.local/bin/uvx
  • Windows: run where uvx in PowerShell and use it with forward slashes, e.g. C:/Users/you/.local/bin/uvx.exe. (Git Bash's which uvx prints /c/Users/..., which Windows doesn't understand: write it as C:/Users/....)
• "This API token is read-only": sign in again and choose Deploy access.
• Deployment built old code: push your commits; DeployKar builds the latest commit on GitHub.
• No build logs: DeployKar keeps build logs for about an hour after a build.

Development

uv run --group dev pytest

Releases are published to PyPI from GitHub (see RELEASING.md in the repository).

License

MIT