For agents
Everything on this site is also data. Connect an agent to the MCP server, install the skill, or read the JSON and Markdown directly. All of it is public and read-only, with no key.
All access
The MCP server
https://aivy.live/mcp speaks the Model Context Protocol over Streamable HTTP. Stateless, read-only, CORS open.
claude mcp add --transport http aivy https://aivy.live/mcp{
"mcpServers": {
"aivy": {
"url": "https://aivy.live/mcp"
}
}
}[mcp_servers.aivy]
url = "https://aivy.live/mcp"npx mcp-remote https://aivy.live/mcpIn the Claude apps: Settings, Connectors, Add custom connector, and paste the URL.
Tools
aivy_canon
Aivy in one call: the eight fixed points, the identity text to paste into image prompts (full for close and medium shots, short for full length and wider, plus the line to use with a reference image), every look's wardrobe text, her voice and manner, and the never-list. Set include_images to get the ids of the canon images (face portrait to use as image 1, sheet, distances).
Arguments: include_images
list_shows
The three Aivy films in the works (CONTEXT WINDOW, HELPFUL, TRELLIS): logline, treatment, the visual device, looks, style, the batch of world-test stills, the motion test, and an honest status per stop. No film is finished yet.
search_gallery
Search and filter every image in the gallery. query matches the prompt, id, note and look (all words must match). Filters: batch (e.g. av-14-looks), look (e.g. request-line), framing (CU, MCU, MS, FS, WS, sheet, insert), film (context-window, helpful, trellis), verdict (pick, reject, note, any; default any). Picks rank first. Returns compact entries; use get_image for the full prompt.
Arguments: query batch look framing film verdict offset limit
get_image
One image by id (batch.job.n, e.g. av-16-helpful.stage.1): the exact prompt and model it ran on, parameters, the verdict and why, its sibling variants, the batch note and URLs at three sizes. Set include_image to receive the picture itself.
Arguments: id include_image size
build_prompt
Compose an image prompt for Aivy the way this site's images were made: framing, the right identity text, the single-tattoo guard on wide shots, wardrobe (a canon look or your own), your scene, a style line and light, ending with no text, no letters, no logos. Returns the prompt, tool parameters and a checklist for the result.
Arguments: scene framing look wardrobe style light reference tool
check_prompt
Check a prompt for Aivy against the canon before you run it: missing identity details, the wrong identity length for the framing, a missing single-tattoo guard on wide shots, green wardrobe, bare arms, the word young, brand or designer names, and requests for lettering. Returns problems with fixes.
Arguments: prompt framing
get_guide
An article of the Aivy wiki as Markdown. Without a topic, lists the articles. Topics: start, consistency, stills, motion, song, storyboard, pipeline, formats. Pass section to get one heading's text.
Arguments: topic section
plan_production
Plan something with Aivy at a given size (still, loop, teaser, full) and song length: shot count, stills to render, clips to generate, credit estimates at the prices measured when this site was made, the decisions a person should keep, and the steps in order.
Arguments: size song_seconds image_model moving_share
get_licence
The licence for Aivy and the images on this site, the three fixed points, notes on releasing music, and what the site cannot grant (Anthropic marks, tool terms, the coin).
get_skill
How to install the Aivy agent skill (canon, wiki, licence and a stdlib-only gallery script, as a zip) and the site's other machine-readable endpoints.
Resources: aivy://canon, aivy://wiki, aivy://licence, aivy://guide/{topic}. Prompts: aivy_still (one image in your scene) and aivy_show (plan a short film, stop for approval before generating).
A first session: aivy_canon, then search_gallery("neon"), get_image(id, include_image=true), build_prompt(scene=…, look="request-line", style="karaoke"), check_prompt on the result.
The skill
The same knowledge as files, for agents that load skills: the canon and rules in SKILL.md, every guide and the licence in references/, and a Python script (standard library only) that searches the gallery and downloads images.
curl -sL https://aivy.live/skill/aivy-skill.zip -o aivy-skill.zip
unzip -o aivy-skill.zip -d ~/.claude/skills/Read it first: /skill/SKILL.md.
llms.txt, Markdown and JSON
llms.txt indexes the site; llms-full.txt is the canon, every guide and the licence in one file. Every page has a Markdown twin: add .md to its path (/aivy.md, /wiki/stills.md, /i/av-16-helpful.stage.1.md).
| Endpoint | What |
|---|---|
/api/index.json | The endpoints and the image fields |
/api/canon.json | The character as data |
/api/shows.json | The three films and their status |
/api/images.json | Every image with its prompt |
/api/images/{id}.json | One image |
/api/guides.json | Every guide as Markdown |
Image files: /img/{batch}/{job}-{n}.s.webp (fits 720 px), .m.webp (fits 1800 px), .full.jpg (native). Please cache what you fetch and prefer the small size.