Documentation

Agent interface

Overview

An agent requests human capability through a small tool surface. The agent specifies the outcome it needs; it does not select a human. The interface is intended to be wrapped as MCP tools once the internal API contract is stable.

Conceptual interface

quote_human()

Request a quote for a task type. Returns availability, price, estimated turnaround and required capability.

quote_human(task_type, inputs, deadline?)

# conceptual response
{
  "available": true,
  "price_usd": 39,
  "estimated_minutes": 30,
  "required_capability": "ux_cro_review"
}

call_human()

Create a task from a valid quote. Returns the task identifier and its initial state.

call_human(quote_id)

# conceptual response
{
  "task_id": "task_…",
  "status": "CREATED",
  "task_type": "LANDING_PAGE_REVIEW"
}

get_result()

Retrieve the structured result once the task is complete.

get_result(task_id)

# conceptual response
{
  "task_id": "task_…",
  "status": "COMPLETED",
  "result": {
    "highest_impact_change": "…",
    "conversion_blockers": ["…"],
    "confidence": 0.85
  }
}

Conceptual task lifecycle

The intended task state machine:

QUOTED → CREATED → OFFERED → ACCEPTED → IN_PROGRESS → SUBMITTED → COMPLETED

Terminal / alternate states: CANCELLED, EXPIRED, REJECTED, FAILED

Planned MCP layer

MCP is planned as a thin adapter over the internal API. Business logic lives in the API layer, not inside MCP. Distribution priority: MCP first, then the ChatGPT / Codex, Claude and Gemini ecosystems, then REST and SDKs.

Planned REST endpoints

Conceptual endpoint shapes, not live URLs:

POST /api/quotes
POST /api/tasks
GET  /api/tasks/:id
POST /api/tasks/:id/accept
POST /api/tasks/:id/start
POST /api/tasks/:id/submit
GET  /api/tasks/:id/result