AAI ProtocolAgent App Interface

From GUI to AAI: Let Agents directly invoke application capabilities

🛡️

Zero Intrusion

Leverages existing platform automation (AppleScript, COM, DBus). No code changes needed for apps that already support automation. Agents connect via standard MCP -- no custom integration required.

⚡

Millisecond Performance

Direct IPC calls complete in 1-10ms, compared to seconds for GUI automation. No screenshots, no OCR, no simulated clicks -- just native inter-process communication.

🌐

Cross-Platform

Unified protocol abstraction with platform-native implementations. macOS (AppleScript/JXA), Windows (COM), Linux (DBus), Web Apps (REST API + OAuth), with a consistent aai.json descriptor format.

What is AAI?

AAI (Agent App Interface) is an open protocol that enables AI Agents to directly invoke application capabilities -- bypassing the slow, fragile approach of "watching screens and simulating clicks."

Traditional:  Agent -> [Screenshot] -> [OCR] -> [Click] -> GUI -> App  (seconds)
AAI:          Agent -> [MCP] -> AAI Gateway -> [IPC/API] -> App         (milliseconds)

Applications provide two independent interfaces:

GUI -- for humans (visual, intuitive)
AAI -- for Agents (structured, programmable, parallel)

Both access the same core business logic. Neither interferes with the other.

How It Works

Apps describe their capabilities in aai.json (placed in ~/.aai/<appId>/aai.json)
AAI Gateway discovers and loads these descriptors
Agents connect to the Gateway via standard MCP (stdio)
Gateway translates MCP requests into platform-native automation calls

Quick Example

json

{
  "mcpServers": {
    "aai": {
      "command": "aai-gateway",
      "args": ["--mcp"]
    }
  }
}

Then ask your Agent:

"Send an email to alice@example.com with subject 'Hello' using Mail"

The Agent calls the Gateway, which executes the operation via native IPC in milliseconds.