For Claude, ChatGPT and any agent

Connect Avsha to your assistant

Avsha turns any place into a short narrated story, grounded in documented history, in 60+ languages. Its library of thousands of ready story pages is available to AI assistants as a read-only tool: search it, fetch a story, or ask for the best story about a place — and get a page with audio and transcript, an MP3, and an “In short” answer to quote with attribution.

MCP connector: avsha-stories

Remote MCP server (Streamable HTTP). No account, no key, no sign-in — it is a public library.

https://avsha.app/mcp

In Claude: Settings → Connectors → Add custom connector → paste the URL. Any MCP client that speaks Streamable HTTP works the same way.

Tools

ToolWhat it doesInputs
search_storiesSearch the library by place, thing or question. Returns ranked stories with page URL, MP3, duration, the “In short” question/answer and a confidence.query, language (ISO 639-1, optional), kind (place | topic, optional), limit (1–25)
get_storyOne story by language + slug (both appear in a story URL: /story/<language>/<slug>/).language, slug
story_for_placeThe best story for a specific place, or — when no page exists yet — an app link that narrates the place on demand. Refuses low-confidence matches instead of guessing.place, language (optional)

All three are read-only, idempotent and non-destructive. They never write, never call third parties, and never see who is asking.

REST + OpenAPI (ChatGPT Actions, any HTTP agent)

GET https://avsha.app/api/stories/search?q=Brooklyn%20Bridge&lang=en&limit=5
GET https://avsha.app/api/stories/for-place?place=Ribeira%2C%20Porto&lang=en
GET https://avsha.app/api/stories/<lang>/<slug>
OpenAPI 3.1: https://avsha.app/openapi.json

Same handlers as the MCP tools. CORS is open, responses are cacheable for five minutes.

Data handling

  • What is sent: only the tool inputs (a search phrase, a place name, a language code). No user identity, no conversation content, no account data.
  • What is stored: nothing per user. Requests appear in ordinary server logs (IP address, path, timestamp) for abuse prevention and are not linked to any account. Rate limit: 120 requests per minute per IP.
  • Where answers come from: a prebuilt index of Avsha’s published story pages — no model is called, so no AI cost and no hallucinated results. If nothing matches, the answer says so.
  • Content: stories are generated by Avsha from documented sources and narrated by an AI voice; each page links to its own transcript. Please attribute quotations to Avsha with the page URL.

Privacy policy: avsha.app/privacy.html · Terms: avsha.app/terms.html

Support

Email partners@avshastories.com. Made by Metivity Ltd, Israel. Status of the library: avsha.app/story/ · machine index: /story/index.json · llms.txt