Agents

Give your agent an image URL.

There is no SDK surface to learn and no key to authenticate, so any model that can write a string can ship imagery. This page hands yours the setup: the files to read, the environment variable to set, and the snippet for the tool you are in.

Your project slug

Every link and snippet below is written for YOUR-PROJECT.staging.imagegpt.host.

Entry points

Everything an agent needs is a URL on this site.

Point yours at one of these. Each one is generated, so it can never advertise a model or a parameter the API does not serve.

entry pointwhat it is

entry point/llms.txt

what it isThe whole URL grammar as a plain-text usage guide. Add ?project=your-slug to bake in your host, or ?lang=python to pick the code examples.

entry point/llms-full.txt

what it isThe same guide with every code example included, for agents with room to read it all.

entry point/SKILL.md

what it isA drop-in skill file for Claude Code and any agent that reads skills, in eleven languages. Takes the same ?project= and ?lang= parameters.

entry point/.well-known/skills/

what it isMachine-discoverable skill index; each entry resolves to the skill file above.

entry pointIMAGEGPT_PROJECT

what it isThe environment variable your agent reads for the project slug — VITE_ and NEXT_PUBLIC_ variants too. Unset, the docs fall back to YOUR-PROJECT.

/llms.txt and /SKILL.md also take ?lang= — one or more of html, react, vue, svelte, javascript, typescript, python, ruby, go, rust and elixir — to keep only the examples you need. Every HTML page here advertises these files in its headers too (X-Llms-Txt and a Link: rel=alternate), so a crawler finds them without reading this page.

Setup

One snippet, whichever tool you are in.

Pick the one you use. Each snippet is written for the project above.

Install the skillone file, once per machine

terminal

export IMAGEGPT_PROJECT=YOUR-PROJECT
mkdir -p ~/.claude/skills/imagegpt
curl -o ~/.claude/skills/imagegpt/SKILL.md "https://imagegpt.cloud/SKILL.md"

Claude Code loads skills from ~/.claude/skills, and /SKILL.md is the same file /.well-known/skills advertises to agents that discover skills on their own. Add ?lang=python (or any of the eleven languages) to narrow the code examples inside it.

What the agent gets back

The response is the image, not a job to poll.

No JSON, no base64, no signed URL that expires. The first request generates the image and every request after it is served from cache, so an agent can write the URL into a component, a README or an email and move on.

Nothing an agent sends can fail the request except a missing prompt: an unrecognised value is replaced with a documented default and reported in an IGPT-Warning header, so a URL written once keeps serving. How invalid parameters behave

request

GET https://YOUR-PROJECT.staging.imagegpt.host/image
    ?prompt=matte+ceramic+mug+on+concrete
    &route=quality/high
    &aspect_ratio=4:3

response

200 image/png

cache-control: public, max-age=31536000, immutable

x-cache: EDGE_HIT

Hand your agent a project of its own.

Free 7-day trial included. Cancel any time.

Generate images Read the docs