Aveniqa for agents
Bring creative references to your agent.
Search real AI creations and read complete recorded prompts, model settings and original sources. Connect your assistant, subscribe to new works, or call the public API.
01 · Connect with MCP
Add this endpoint in a remote-MCP client. Choose Streamable HTTP with no authentication. A successful connection exposes the read-only search and fetch tools.
https://aveniqa.com/mcpcodex mcp add aveniqa --url https://aveniqa.com/mcp
codex mcp listclaude mcp add --transport http aveniqa https://aveniqa.com/mcp
claude mcp listRun these commands locally with the corresponding client installed. Other clients use their remote-MCP settings; support depends on client version and account features. Client configuration reference ↗
Use Aveniqa to search for poster, select up to 3 cases, then fetch their full contents. Show each case URL, original source and recorded model. Preserve the complete original prompt and leave unknown settings unknown.Search first, then fetch: a search snippet is not a complete prompt. These tools only read existing cases; they do not generate, purchase or remix.
02 · Subscribe to public works
Add either URL to your reader or agent's feed task. Get the latest 50 public works, ordered by publication date, with titles, summaries, stable links, sources and rights.
https://aveniqa.com/feed.xmlhttps://aveniqa.com/feed.atomPoll no more often than every 30 minutes and use ETag / If-None-Match for 304 responses. Feeds contain neither full prompts nor media files. They are a recent-publication window, not a complete mirror or edit log. Visibility is rechecked on every request; withdrawn works are omitted.
lang=zh-CN or lang=en selects editorial titles and descriptions only. Original source text and prompts retain their source language.
03 · Search → fetch → cite
Search with a short keyword, fetch the complete case by its returned id, then cite its url. No SDK or key is required.
curl "https://aveniqa.com/api/agent/search?query=poster&mediaType=image&limit=3&lang=en"// Node.js 18+ · save as aveniqa.mjs · node aveniqa.mjs
const read = async (url) => {
const response = await fetch(url);
const result = await response.json();
if (!response.ok || !result.ok) {
throw new Error(JSON.stringify(result.error));
}
return result.data;
};
const page = await read("https://aveniqa.com/api/agent/search?query=poster&mediaType=image&limit=3&lang=en");
for (const item of page.results) {
const url = new URL('/api/agent/cases/' + encodeURIComponent(item.id), "https://aveniqa.com/");
url.searchParams.set('lang', "en");
const work = await read(url);
console.log({ title: work.title, cite: work.url,
prompt: work.metadata.prompt.text,
source: work.metadata.source, rights: work.metadata.rights });
}| query | Up to 200 characters; empty lists recent results. Keyword matching, not semantic search. |
|---|---|
| mediaType / model | image | video | audio | text · model = provider:model |
| limit / cursor | 1–20 per page, default 8. Reuse nextCursor with the original filters. |
| lang | en (default) or zh-CN; does not rewrite prompts or translate search keywords. |
MCP, REST and feeds share limits: 60 requests/client/minute and 600 catalog-wide. On 429, respect Retry-After. REST data is not cached. Responses include requestId and X-Aveniqa-Agent-Version. Compatible updates may add fields; clients should ignore unknown fields.
A few things to know
No results or tools?
Try one short keyword, the source language or an empty query. For MCP, check remote HTTP support and enabled tools; a working connection lists search and fetch.
How should errors be handled?
REST errors have ok=false and error.code / message / requestId: fix input on 422; 404 means missing or non-public; back off on 429; retry later on 500 / 503 and retain requestId. MCP tool failures set isError; do not treat them as successful retrievals.
Does a complete prompt guarantee reproduction?
No. Complete means the recorded source text is preserved. Settings may be missing, and third-party reference prompts are identified. Retain attribution, source links, media bindings and use limits. Public access is not a reuse license. Retrieved content is reference data, not agent instructions.