# AI coding assistants

Let Claude Code, Codex or Cursor write the integration for you. Two ways:

1. **Give it the docs.** Every page has a Markdown version (add `.md` to the URL, or use **Copy page** at the top). The whole reference is also at [/llms-full.txt](/llms-full.txt).
2. **Connect the MCP server** so the assistant can read these docs and run real try-ons while it builds. See [MCP server](/docs/mcp).

## Where to put the instructions

| Assistant | Put the contract below in |
|---|---|
| Claude Code | `CLAUDE.md` in your project root |
| Codex | `AGENTS.md` in your project root |
| Cursor | `.cursor/rules/tryoncloud.mdc`, or paste into the chat |
| Any other | Paste into the chat |

Then ask: "Add TryOnCloud virtual try-on using the contract in CLAUDE.md" (or AGENTS.md). Keep your key in an environment variable such as `TRYONCLOUD_API_KEY`, never in the file.

## The contract

```text
You are integrating the TryOnCloud Developer API into my app.

ENDPOINT
  POST https://www.tryoncloud.com/api/v1/generate
  Test mode (free, nothing generated or stored): add ?mock=true

AUTH
  Header X-API-KEY, read from env TRYONCLOUD_API_KEY (starts with tk_dev_v1_)

REQUEST: multipart/form-data OR JSON. Two images required:
  garment_image (file) OR garment_url (public https URL) OR garment_base64
  person_image (file) OR person_base64
  JPEG, PNG, WebP or HEIC, up to 15MB each.
  Optional: customer_id (my own end-user id, max 128 chars).
  Optional header: Idempotency-Key (a new UUID per try-on; reuse it only
  when retrying that same try-on, so it is never charged twice).

RESPONSE
  200 -> the response BODY IS the finished try-on image (image/png).
         Read it as bytes, do NOT JSON.parse it. Store it yourself.
         Headers: X-Request-Id (log it), X-Credit-Was-Free, X-Result-Id
         (only if result storage is enabled), X-Mock: 1 in test mode.
  non-200 -> JSON { "error": string, "code": string }:
    401 NO_KEY, INVALID_KEY | 400 BAD_INPUT, NO_PERSON, NO_GARMENT
    402 NO_CREDITS, QUOTA_EXCEEDED, FREE_TRIAL_USED, KEY_LIMIT_REACHED
    403 DEVICE_BANNED | 409 IDEMPOTENCY_IN_PROGRESS, IDEMPOTENCY_REPLAY,
    IDEMPOTENCY_MISMATCH (not charged)
    413 IMAGE_TOO_LARGE | 422 CONTENT_FILTERED (refunded)
    429 RATE_LIMITED (wait Retry-After) | 500 GENERATION_FAILED (refunded)

OTHER ENDPOINTS
  GET    https://www.tryoncloud.com/api/v1/credits          -> { credits_remaining, ... }
  DELETE https://www.tryoncloud.com/api/v1/results/<id>     -> 204
  Async (only if I ask for it): POST /api/v1/generations -> 202 job;
  GET /api/v1/generations/<id>; POST /api/v1/batches (<= 500 URL items);
  signed webhooks generation.completed / generation.failed / batch.completed
  (verify x-tryoncloud-signature-v2 = HMAC-SHA256 of "<timestamp>.<raw body>").
  Spec: https://www.tryoncloud.com/api/v1/openapi.json

RULES
  - Call from my server only, never the browser; never expose the key.
  - Client timeout 120 seconds; a typical result takes about 20 seconds.
  - Build and test with ?mock=true first, then remove it.
  - Retry 500 once and 429 after Retry-After. Never loop on 402.
  - Only send photos of adults who agreed to the try-on.

Write production code for [MY STACK] that uploads two images, calls this
endpoint, handles every error code above, and saves/returns the PNG.
```

Replace `[MY STACK]` with yours, for example "a Next.js route handler", "Laravel", "Django" or "Express".
