# ontology.guru

> 120 published ontologies, read and catalogued by hand. 50 of them serve a class hierarchy that was extracted from the ontology's own published source file; the other 70 deliberately serve none, and say why. Free JSON API over the lot — no key, no rate limit, CORS open.

Read this before you use the data. Two fields decide whether an entry can be trusted:

- `hierarchy.status` is `verified` (parsed out of `hierarchy.source` on `hierarchy.retrievedAt`, by `hierarchy.extractionCommand`) or `withheld` (no hierarchy is served; `hierarchy.reason` says why). **A withheld entry does not have an empty hierarchy — it has an unpublished one.** Do not report "0 classes" for one.
- `licence.verified` is `true` for only 26 of 120 entries (22%). For the rest, the licence label is the catalogue's original guess and was never checked against the licence text. Say so when you pass it on.

Verified hierarchies are **pruned subsets chosen to be readable, not complete ontologies**. `hierarchy.classesShown` versus `hierarchy.classesInSource` tells you how much was left out. Never present one as the full ontology.

Licence summaries here are made in good faith from primary sources and are not legal advice. Check the licence yourself before you redistribute anything.

## API

- [Catalog](https://ontology.guru/api/catalog.json): every entry — metadata, licensing with the operative terms, and hierarchy status. One fetch, no pagination.
- [Single entry](https://ontology.guru/api/ontology/prov-o.json): one ontology, plus its class hierarchy when verified, a worked SPARQL example, and sample instance data. Pattern: `/api/ontology/<id>.json`.
- [Verified ids](https://ontology.guru/api/verified.json): just the ids whose hierarchy is source-verified. Filter on this before you trust a tree.
- [Recommender](https://ontology.guru/for-agents): `POST /api/recommend` with `{"prompt": "what you are modelling"}`. Backed by a language model on a personal key — best-effort, and it can be down.

## MCP

- [MCP server](https://ontology.guru/mcp): a remote [Model Context Protocol](https://modelcontextprotocol.io) server over stateless Streamable HTTP. POST JSON-RPC to `https://ontology.guru/mcp`; no key, no session, no auth. 5 read-only tools:

- `search_ontologies` — Search 120 published ontologies by keyword, domain, licence, or whether their class hierarchy was verified against the published source file. Start here — it is deterministic and free. Returns matching entries with licence and hierarchy status.
- `get_ontology` — Full record for one ontology by id, including its class hierarchy when that hierarchy was extracted from the published source file. Ids come from search_ontologies — never guess one.
- `check_licence` — The operative licence terms for one ontology: whether you may redistribute it, whether attribution is required and with exactly what notice, and whether share-alike propagates to anything you derive from it. Says plainly when the licence was never verified.
- `list_verified` — Every id whose class hierarchy was extracted from the ontology's own published file. The cheapest way to filter before trusting a tree.
- `recommend_ontologies` — Describe what you are modelling in prose and get suggested ontologies — often a combination — with a reason for each. This one calls a language model, so it is slower, best-effort, and can be unavailable. Prefer search_ontologies when you already know the domain or keywords; use this when the problem is easier to describe than to search.

## For agents

- [Full instructions](https://ontology.guru/llms-full.txt): field-by-field reference, the licence and withheld vocabularies, worked recipes, and the complete 120-entry index. Fetch this one if you are going to do more than a single lookup.
- [Agent skill](https://ontology.guru/skill.md): the same guidance as an Agent Skill file, ready to drop into `.claude/skills/ontology-guru/SKILL.md`.

## Optional

- [Human API docs](https://ontology.guru/for-agents): the same endpoints, written for a person.
- [Homepage](https://ontology.guru/): describe what you are modelling and get a recommendation; browse the catalogue by domain.