Can I use CelluloidScore data for AI training or retrieval?

Yes. Our robots.txt carries Content-Signal directives permitting search, ai-input and ai-train. We ask for attribution — link the specific title URL and note the "Last reviewed" date.

Do I need an API key to download the data?

No. Every bulk file is public and needs no authentication. Only the interactive MCP and A2A endpoints require a token, and registration is open — a single POST with no approval step.

How do I get Markdown instead of HTML?

Send an Accept: text/markdown request header, or append .md to any film or series URL. Both return the same fact sheet.

How often does the data change?

Scores change as new reviews land, and the catalogue grows continuously. Each fact sheet carries its own Last reviewed date, and the sitemap carries per-URL lastmod dates drawn from content history.

Data & AI Access | Celluloid

Bulk downloads

File	Format	Contents
Endpoint manifest	`application/json`	The canonical index of every machine-readable endpoint on the site.
Site summary for language models	`text/plain`	Section map, coverage figures and citation guidance. Generated at build time, so the numbers are never stale.
Passage chunks (NDJSON)	`application/x-ndjson`	Answer-sized passages for embedding and retrieval pipelines, one JSON object per line.
Passage chunks (single bundle)	`application/json`	The same passages as one JSON array, with cl100k_base token counts.
Title and people search index	`application/json`	Lightweight index of every film, series, review and star profile with its canonical URL.

Per-title fact sheets

Every film and series has a plain Markdown fact sheet at the same path with a .md suffix — for example /movies/12-angry-men-1957.md. Each one carries the Celluloid Score breakdown, critic consensus, cast, synopsis and streaming availability.

You can also request Markdown by content negotiation: send Accept: text/markdown and the homepage and any film or series page will return Markdown instead of HTML.

Interactive access for agents

When you need search and filtering rather than the whole corpus, two authenticated endpoints expose the same catalogue tools — search_catalog, get_factsheet and list_top_rated:

/api/mcp — Model Context Protocol over Streamable HTTP
/api/a2a — A2A agent endpoint (JSON-RPC message/send)

Both use OAuth 2.0 with open dynamic client registration. The full walkthrough, including the token request, is at /auth.md.

Licence and attribution

The dataset is offered under CC BY 4.0. When you cite a score or review, link the specific title page rather than the site root, and include the Last reviewed date shown on it — scores change as new reviews land. Attribute review text to the named author in the byline. Our scoring policy explains how the composite is built.