MCP server

@descanto/mcp is an MCP (Model Context Protocol) stdio server exposing Descanto cloud desktops to MCP-compatible AI clients -- Claude Code, Claude Desktop, Cursor, Codex, GitHub Copilot, Antigravity, OpenCode, and any other MCP client. It's a thin wrapper over @descanto/sdk: one process, one org's worth of desktops, driven entirely through the tools below.

Configuration

Claude Code

claude mcp add descanto --env CANTO_API_KEY=canto_sk_... -- npx -y @descanto/mcp

Claude Desktop

Add to claude_desktop_config.json (Settings → Developer → Edit Config):

{
  "mcpServers": {
    "descanto": {
      "command": "npx",
      "args": ["-y", "@descanto/mcp"],
      "env": {
        "CANTO_API_KEY": "canto_sk_..."
      }
    }
  }
}

Cursor

Add to .cursor/mcp.json (project) or ~/.cursor/mcp.json (global):

{
  "mcpServers": {
    "descanto": {
      "command": "npx",
      "args": ["-y", "@descanto/mcp"],
      "env": {
        "CANTO_API_KEY": "canto_sk_..."
      }
    }
  }
}

Codex

Use the CLI:

codex mcp add descanto --env CANTO_API_KEY=canto_sk_... -- npx -y @descanto/mcp

or add the server to ~/.codex/config.toml by hand:

[mcp_servers.canto]
command = "npx"
args = ["-y", "@descanto/mcp"]
env = { CANTO_API_KEY = "canto_sk_..." }

GitHub Copilot (VS Code)

Add to .vscode/mcp.json. Copilot uses the top-level servers key and requires "type": "stdio":

{
  "servers": {
    "descanto": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@descanto/mcp"],
      "env": {
        "CANTO_API_KEY": "canto_sk_..."
      }
    }
  }
}

Antigravity

In the agent panel, open … → MCP Servers → Manage MCP Servers → View raw config (writes ~/.gemini/config/mcp_config.json):

{
  "mcpServers": {
    "descanto": {
      "command": "npx",
      "args": ["-y", "@descanto/mcp"],
      "env": {
        "CANTO_API_KEY": "canto_sk_..."
      }
    }
  }
}

OpenCode

Add to opencode.json (project) or ~/.config/opencode/opencode.json (global). OpenCode nests servers under mcp, takes command as an array, and uses environment rather than env:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "descanto": {
      "type": "local",
      "command": ["npx", "-y", "@descanto/mcp"],
      "enabled": true,
      "environment": {
        "CANTO_API_KEY": "canto_sk_..."
      }
    }
  }
}

Tools

fork_desktop requires acknowledge_shared_state: true explicitly -- omitting it is rejected the same as passing false, since a fork duplicates whatever in-guest state (including credentials) was live in the source desktop's memory when it last hibernated. See Concepts: Forking.

Computer-use tools (screenshot, click, move_mouse, drag, scroll, type_text, press_key, set_resolution) require an awake desktop and are never gated by a noVNC control session -- a human viewer and these tools can interleave. They're also not transactional: an error mid-batch means earlier effects already applied, so re-screenshot before retrying rather than blindly re-sending clicks. See Concepts: Computer use.

expose_port/list_ports/stop_port mirror POST/GET/DELETE .../ports directly -- see Concepts: Port ingress for the token/hibernation model.

Output truncation (context protection)

exec's stdout/stderr are each truncated at 50KB in the tool result (never in the underlying API call) with a trailing [truncated by canto-mcp] marker, UTF-8-boundary-safe. This exists purely to protect the calling model's context window -- for larger output, redirect the command's output to a file on the desktop and fetch it with read_file, or re-run with head/tail.

read_file follows the same idea: content is returned as text only when it's valid UTF-8 and under 50KB; otherwise it's base64-encoded, but only if that encoding itself stays under 64KB (base64 inflates size ~4/3x, so this in practice means small binary files only) -- otherwise the tool returns { size, error: "too large for tool result — use exec with head/tail" }.

Errors

A failed Descanto API call never crashes the server. Every tool catches CantoApiError (and any other thrown error) and returns an MCP tool-error (isError: true) whose message includes the API's HTTP status, title, and detail (and operation_id when the problem concerns a specific operation), so the calling model sees exactly what the API returned and can react to it.