Set up your client

Pick your AI client below, paste the config, and you're connected. Every snippet uses MC_TOKEN as a placeholder — substitute the API token you copied from vm.massedcompute.com/settings/api.

Claude Code

The fastest path is the CLI:

claude mcp add --transport http massed-compute https://vm.massedcompute.com/api/mcp --header "Authorization: Bearer MC_TOKEN"

This registers the server globally. Run claude from any project and ask "list my Massed Compute instances" — Claude Code discovers and calls the tools automatically.

To configure by hand instead, edit ~/.claude.json (global) or .mcp.json in a project root:

{
  "mcpServers": {
    "massed-compute": {
      "type": "http",
      "url": "https://vm.massedcompute.com/api/mcp",
      "headers": {
        "Authorization": "Bearer MC_TOKEN"
      }
    }
  }
}

Useful commands: claude mcp list shows registered servers, claude mcp remove massed-compute uninstalls, and claude --debug traces tool list/call activity if something looks off.

Cursor

Edit ~/.cursor/mcp.json (global) or .cursor/mcp.json in a project. Cursor watches both:

{
  "mcpServers": {
    "massed-compute": {
      "url": "https://vm.massedcompute.com/api/mcp",
      "headers": {
        "Authorization": "Bearer MC_TOKEN"
      }
    }
  }
}

Open Settings → Cursor Settings → MCP. The massed-compute server should show with a green status dot. A red dot usually means a bad token, an old Cursor build (HTTP MCP support requires Cursor 0.45+; older builds only support stdio), or a network block — hover the dot for the specific error.

In chat, type @massed-compute to inline the tool list, or just describe what you want ("show my running VMs") and Cursor will route to the appropriate tool.

Codex

Edit ~/.codex/config.toml:

[mcp_servers.massed-compute]
url = "https://vm.massedcompute.com/api/mcp"
http_headers = { Authorization = "Bearer MC_TOKEN" }
enabled = true

For better token hygiene, source the token from an environment variable instead of putting it in the config file:

[mcp_servers.massed-compute]
url = "https://vm.massedcompute.com/api/mcp"
bearer_token_env_var = "MC_TOKEN"
enabled = true

Then add the token to your shell profile:

export MC_TOKEN="paste-your-api-token-here"

Restart Codex. Tools should appear via /tools or surface naturally when you describe a task.

Claude Desktop

Claude Desktop speaks MCP over stdio, so use the mcp-remote bridge to proxy HTTP. Edit claude_desktop_config.json:

{
  "mcpServers": {
    "massed-compute": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://vm.massedcompute.com/api/mcp",
        "--header",
        "Authorization: Bearer MC_TOKEN"
      ]
    }
  }
}

Config file paths: macOS at ~/Library/Application Support/Claude/claude_desktop_config.json, Windows at %APPDATA%\Claude\claude_desktop_config.json.

Restart Claude Desktop after saving. Tools appear in the bottom-right tool indicator.

Verify you're connected

Once you've saved your client config, ask your assistant something simple like "Validate my Massed Compute API key" or "List my running Massed Compute VMs." If it responds with real data, you're wired up.

If the assistant says it doesn't have access, or you get an authentication error, see Troubleshooting.

What's next

• Tools reference — what your AI assistant can do, with example prompts
• Agent Skills — downloadable workflow guides for picking GPUs, launching VMs safely, auditing spend
• Troubleshooting — common issues and fixes