Integrations Agents

AI Agents

Deskbrid v1.0.0 exposes desktop control to AI coding agents through the MCP server, or directly by sending actions over the daemon socket. The daemon is the source of truth for available actions; see API.md for the canonical action map.

MCP server

Stdio mode (AI tools)

deskbrid daemon
deskbrid mcp

Clients connect over stdin/stdout.

TCP mode

deskbrid daemon --mcp-port 20129

Clients connect to tcp://localhost:20129.

Client configs

Claude Desktop:

{
  "mcpServers": {
    "deskbrid": {
      "command": "deskbrid",
      "args": ["mcp"]
    }
  }
}

Cursor:

{
  "mcpServers": {
    "deskbrid": {
      "command": "deskbrid",
      "args": ["mcp"]
    }
  }
}

Agent safety

• Elevated or sensitive actions flow through the dashboard challenge / confirmation UI unless explicitly allowed by policy.
• Use the require_confirmation / confirmation flow to review uncertain agent actions before dispatch.
• Policy enforcement mirrors socket authentication; the safer path is via confirm.challenge / confirm.resolve, not by passing extra permissions over the wire.

Audit trail

Review agent actions via the dashboard or audit log:

deskbrid audit_log