Developer Platform · Beta
Image, video & tools API
The same engine that powers the Caranguejo app, now in your code. Dozens of image and video models, plus tools like upscale, relight and inpaint — via REST, SDK, CLI, an MCP server, and Skills for agents.
Every generation goes through a single asynchronous endpoint. You send a model (the slug of a model or tool) plus the prompt and parameters, get back a processing job, and fetch the result by polling or webhook. The foundation for everything:
# Gere uma imagem em uma linha
curl -X POST https://caranguejo.art/api/v1/dev/generations \
-H "Authorization: Bearer $CK" -H "Content-Type: application/json" \
-d '{ "model": "gpt-image-2", "prompt": "um caranguejo numa praia neon", "params": { "quality": "high" } }'Start with the Quickstart to generate your first image in minutes.
Four ways to integrate
From plain REST to AI agents — pick the surface that fits you. All of them use the same key and the same credits.
REST API
Asynchronous endpoints with signed webhooks, uploads, and cursor-based pagination.
MCP Server
Connect Claude, Cursor, and other agents. Generate via tool-calling in seconds.
CLI
Generate straight from the terminal or from scripts. --wait and --json included.
Agent Skills
Ready-made playbooks that talk directly to the MCP server — /caranguejo:generate.
Everything you can generate
A living list of models and tools. See Models or GET /models for the full, up-to-date catalog.
Image image
Text-to-image, editing, and reference (multi-image). Low/medium/high quality, up to 4K, transparent background, multiple formats.
Video video
Text-to-video and image-to-video with control over duration, aspect ratio, and start/end frames. Several state-of-the-art engines.
Tools tool
Post-production: upscale, relight, enhance, and inpaint on images or video — no prompt, just the input media.
POST /generations: just swap out model and params. Credit usage comes back in credits_charged.