For AI agents

If an AI agent (Claude, ChatGPT, Cursor, your own) is going to use this API, everything it needs is published in machine-readable form.

Sources

Address	Content
/llms.en.txt	A short index: rules, guides, every endpoint on one line
/llms-full.en.txt	The guides and every endpoint in detail (schema, errors, examples)
/openapi.en.json	OpenAPI 3.1 — for generating tool definitions
/md/<page>.md	Each guide as plain Markdown

Turkish versions: /llms.txt, /llms-full.txt, /openapi.tr.json.

Every operation in the OpenAPI document carries these extensions:

• x-ug-role — which key role may call it (customer, usta, any)
• x-ug-scope — the scope it needs (read, write, money)
• x-ug-error-codes — the error codes it can return
• x-ai-hint — a short hint for agents, where useful

A ready prompt

Add this to your agent's system prompt:

You are using the UstaGeliyor Partner API. Description: <DOCS>/llms-full.en.txt
Base URL: <API>/partner/v1. Send "Authorization: Bearer <KEY>" on every request.
Rules:
- Take service and package ids from GET /catalog; never send prices or titles.
- Show the customer the amount from POST /pricing/preview.
- On errors, decide by the "code" field; you may relay the message to the user.
- Call operations that need the "money" scope (opening a payment, confirming
  a job, refunds, disputes) only after the user explicitly agreed.
- Never ask for or send card details; give the user payment.payment_url.
- Follow state changes with GET /notifications/sync.

Security

• Give the key to the tool the agent calls (a server-side function), not to the model itself; the key should never appear in model output.
• Ask for the least-privileged key: read if the agent only informs.
• Every call shows in the team's log under the key's name; a separate key for the agent keeps its actions apart.