Skip to content

Connecting your AI assistant

Your server offers an MCP server: a toolbox with which an AI assistant can operate your environment. Which steps are needed depends on the assistant.

Assistant Signs in with See
Claude Code (command line) token below
Cursor, VS Code, Windsurf and other editors token below
Claude Desktop (the Windows/Mac app) token, through a bridge below
claude.ai and ChatGPT (the web versions) signing in with your account OAuth
Your own scripts and pipelines token The REST API

The Integrations > Connect your AI tab in the portal shows the same configurations, already filled in with your server’s address.

The Connect your AI tab

For every variant except claude.ai and ChatGPT you need a token. Create it on Integrations > API tokens with the recommended scopes; see API tokens and scopes. The address of the MCP server is:

https://<your-server>.hddeploy3000.nl/mcp

One command in a terminal:

claude mcp add --transport http hddeploy https://<your-server>.hddeploy3000.nl/mcp --header "Authorization: Bearer hdd_..."

Check with claude mcp list; hddeploy should say “Connected”. Test with the question “Who am I on my HDDeploy server and which applications are running?”.

Cursor, VS Code, Windsurf and similar editors

Section titled “Cursor, VS Code, Windsurf and similar editors”

Add an MCP server of type “http” (sometimes called “streamable HTTP”). In most editors that is a JSON file with MCP servers:

{
"mcpServers": {
"hddeploy": {
"type": "http",
"url": "https://<your-server>.hddeploy3000.nl/mcp",
"headers": {
"Authorization": "Bearer hdd_..."
}
}
}
}

Claude Desktop connects directly only to MCP servers through OAuth, and indirectly through the mcp-remote bridge, which the app starts locally. This is a one-time setup of about five minutes.

Before you start: Claude Desktop (the app, not the website), Node.js (LTS version from nodejs.org, default installation; this provides the npx command) and a token.

  1. Open Claude Desktop, go to Settings > Developer > Edit Config. The folder with claude_desktop_config.json opens (%APPDATA%\Claude on Windows).
  2. Open that file in a text editor. If it is empty or does not exist, paste the whole text below. If it already has an mcpServers block, add only the hddeploy part to it.
  3. Replace <your-server>.hddeploy3000.nl with the address of your environment and hdd_... with your token (mind: the word Bearer, a space, then the token).
  4. Save and quit Claude Desktop completely (also from the tray icon: Quit). Start the app again.
  5. On first use npx downloads the bridge; that can take half a minute. After that the chat window shows the tools icon with “hddeploy” and the list of tools.
{
"mcpServers": {
"hddeploy": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"https://<your-server>.hddeploy3000.nl/mcp",
"--header",
"Authorization:${AUTH_HEADER}"
],
"env": {
"AUTH_HEADER": "Bearer hdd_..."
}
}
}
}

Not working?

  • No tools icon: check that Node.js is installed (node -v in a command prompt), that the JSON file is valid (every comma and brace counts) and that Claude Desktop was really quit completely and restarted.
  • “Unauthorized” or “401”: the token is wrong, expired or revoked, or bound to networks and you are on another address.
  • Claude Desktop’s logs are in %APPDATA%\Claude\logs; the file mcp-server-hddeploy.log shows what the bridge reports.

Alternatively, Claude Desktop can be connected as a “custom connector”, like claude.ai: you then sign in with your account and need neither a token nor a bridge. See Signing in from claude.ai and ChatGPT.

The web versions cannot carry a token; they sign in through your browser with your own account. The steps are on Signing in from claude.ai and ChatGPT.

First ask the assistant who it is and what is running. It then calls whoami and list_stacks. After that you can give instructions like those in Example prompts. Read Working safely with an AI assistant first.