Public API, version 1

The university documents for your own AI model

Give a model one address and it can search the official documents of Óbuda University: regulations, curricula, subject sheets, timetables, the academic calendar and procedures. It receives the passages that answer its question, with their sources and an evidence decision, and answers in its own conversation. The API retrieves and never generates. No login, no key, read-only.

Base URLhttps://uni-context-agent.iditechs.com

Context for a question

GET /api/v1/context?q=… returns numbered context blocks with document, version, section, page, validity window, tags and URL, the list of sources, and the evidence decision. When the evidence is insufficient, no passage is returned, only the closest documents.

curl "https://uni-context-agent.iditechs.com/api/v1/context?q=Meddig%20lehet%20vizsg%C3%A1ra%20jelentkezni%3F"

The same answer as Markdown, for a model that reads text:

curl -H "Accept: text/markdown" "https://uni-context-agent.iditechs.com/api/v1/context?q=When%20does%20the%20exam%20period%20start%3F&lang=en"

Parameter	Meaning
`q`	The question, at most 500 characters, Hungarian or English.
`lang`	`hu` or `en`, detected when absent. Both languages are searched either way.
`programme`	A programme code, `cse-bsc` by default.
`type`	Document types, comma-separated: regulation, curriculum, subject_sheet, timetable, calendar, procedure.
`tags`	Tag slugs, comma-separated; a document qualifies when it carries any of them (see /api/v1/tags).
`date`	The day the answer must be valid for, `YYYY-MM-DD`; today by default.
`budget`	Token budget of the context, 3000 by default, at most 6000.
`format`	`json` (default) or `markdown`.

MCP server

Add the endpoint below to any client that speaks the Model Context Protocol over Streamable HTTP. It offers two read-only tools, search_university_documents and list_university_documents, with input and output schemas.

MCP endpointhttps://uni-context-agent.iditechs.com/api/mcp

A client configuration that names it:

{
  "mcpServers": {
    "uni-context-agent": {
      "type": "http",
      "url": "https://uni-context-agent.iditechs.com/api/mcp"
    }
  }
}

Discovery

/llms.txt: what the corpus covers, when it was last updated and how to call the API.
/openapi.json: the OpenAPI 3.1 document, for frameworks that import tools from it.
/api/v1/documents: the catalogue, filtered by type, lang, programme and tags, with cursor pagination.
/api/v1/tags: the tag library.

Limits and errors

Each IP address may send 20 a minute and 500 a day, and the whole API makes at most 2000 model calls a day. Every answer announces the limits in the RateLimit-Policy and RateLimit header fields and carries an X-Request-Id. Identical requests are answered from a cache for ten minutes. The question text is not stored.

Errors are problem details (RFC 9457, application/problem+json) with these stable types:

invalid-request 400: A parameter is missing or invalid; the errors field names each one.
not-found 404: No served document has this id.
method-not-allowed 405: The MCP endpoint is stateless and takes POST requests only.
rate-limited 429: The limit of your IP address is reached; retry after the seconds of Retry-After.
daily-model-cap 429: The daily number of model calls of the whole API is reached; cached answers are still served.
service-unavailable 503: Retrieval is not available at the moment, for example while an index is rebuilt.
internal-error 500: The request failed on the server; retry later.