theses-mcp
MCP server that enables searching and retrieving Czech university theses from theses.cz, including full-text search, metadata, and direct PDF downloads with optional cookie-based authentication.
README
theses-mcp
MCP server for theses.cz — the Czech national registry of university theses (~1M records from most Czech universities, operated by the Faculty of Informatics, Masaryk University).
Gives any MCP client full-text search over Czech bachelor/master/doctoral theses, their metadata, and direct links to the publicly readable PDFs.
Tools
| Tool | What it does |
|---|---|
search(query, limit=10) |
Search records and thesis full texts. Supports theses.cz operators (AND, OR, "phrase"). Paginates by 10, max 50. |
detail(id_or_url) |
Full record: author, CS/EN title and abstract, keywords, supervisor and opponent, defense date, full-text availability, school archive link, related theses. |
fulltext(id_or_url) |
Follows the record into the school's own repository and lists the downloadable files — the thesis PDF plus supervisor/opponent reports. |
search returns two kinds of hits:
kind="record"— a catalogue record;urlpoints to the detail pagekind="fulltext"— a match inside the thesis PDF;pdf_urlis a direct link to the full text, which your client can read directly
Getting the PDF
theses.cz stores metadata only — the files live in each school's own system. 64
institutions are registered, but they are not 64 different problems: the large majority run
the same IS software as theses.cz itself (is.muni.cz, is.slu.cz, is.ambis.cz,
is.vsfs.cz, is.jamu.cz, …), with vskp.vse.cz and a few others alongside. fulltext
follows the record into whichever one it is and lists the documents on offer:
{
"access": ["světu"],
"archive_url": "https://is.muni.cz/th/avwwh/",
"files": [
{"label": "Plný text práce", "url": "https://is.muni.cz/th/avwwh/Moutelik_Bakalarska_Prace_Final.pdf"},
{"label": "Posudek vedoucího", "url": "https://is.muni.cz/th/avwwh/posudek_vedouciho_Minjarikova.pdf"}
]
}
access does not decide what you can download. It describes the copy held by
theses.cz, not the school's own policy. The VŠE thesis pl09jx is marked "autentizovaným
zaměstnancům ze stejné školy/fakulty" on theses.cz while vskp.vse.cz links its 7.2 MB
PDF to anyone — so always look at files, not at access.
That link had no .pdf in the URL; the filename was only in the link text. Files are
therefore matched by href and by label, and extensionless URLs are confirmed with a
HEAD request (confirmed, content_type) rather than optimistically reported as PDFs.
What actually comes back
Measured over ~500 records from 20 unrelated subject searches, covering 17 distinct repository hosts:
| repository | schools | result |
|---|---|---|
is.muni.cz |
Masaryk University | works — full text, both reviewer reports, often DOCX and TXT |
vskp.vse.cz, www.vse.cz |
VŠE Praha | works — thesis, appendix, both reports (extensionless /zp/ URLs) |
hdl.handle.net |
DSpace repositories, e.g. VŠB-TUO | works — handle redirects to the repository, bitstreams confirmed |
is.ambis.cz, is.vsfs.cz, is.slu.cz, is.caritas-vos.cz, is.vstecb.cz, is.vszdrav.cz, is.cevro.cz, is.ucp.cz, is.jamu.cz, is.jabok.cz |
AMBIS, VŠFS, SU Opava, CARITAS, VŠTE, VŠ zdravotnická, CEVRO, UCP, JAMU, JABOK | CAPTCHA for anonymous visitors |
is.vsh.cz, is.vshe.cz |
VŠH, VŠHE | broken TLS certificate (hostname mismatch) |
evskp.uhk.cz |
Hradec Králové | different system, not implemented |
stag.upol.cz, portal.ujep.cz |
UPOL, UJEP | STAG portal, not implemented |
theses.cz lists 64 participating institutions; roughly twenty never appeared in the sample, so their repositories are untested rather than known-good.
The CAPTCHA is not bot detection — is.muni.cz serves this same client fine. Those
schools gate anonymous access, and a human in a browser meets the same wall. This server
reports it and does not try to defeat it.
Using your own login
The way past a gate is to be someone entitled to pass it. Theses marked "všem autentizovaným" (any authenticated user) and repositories that gate anonymous visitors both open up once you are logged in. Log in through your browser, copy the session cookie and hand it over per host:
THESES_COOKIES='{"theses.cz": "__Host-issession=…", "is.vsfs.cz": "__Host-issession=…"}'
Cookies are scoped to their own domain and are never sent to any other host. This gets you what your account is entitled to — nothing more.
Known gaps
- STAG portals (
stag.upol.cz, same software at ZČU, UPCE, JČU, TUL) put the download behind a session-bound portlet flow withpc_phsand_csrfparameters, and UPOL does not expose the STAG web service at the usual/ws/services/rest2/path. evskp.uhk.czandportal.ujep.czare their own systems, unparsed.is.vsh.cz,is.vshe.czserve certificates that do not match their hostname. The session verifies TLS and so refuses them; turning verification off would fix the symptom and remove the guarantee that you are talking to the school at all.
PRs welcome.
Install
Claude Code:
claude mcp add theses -s user -- uvx --from git+https://github.com/koprjaa/theses-mcp theses-mcp
Any other client, in mcpServers:
{
"theses": {
"command": "uvx",
"args": ["--from", "git+https://github.com/koprjaa/theses-mcp", "theses-mcp"]
}
}
From a clone, use pip install -e . and command: "theses-mcp".
Example
> find Czech theses about Midjourney and copyright
search("midjourney AND autorství") → 12 hits, incl. public PDFs
detail("c5uqln") → abstract, keywords, supervisor, defense date
fulltext("7lfo74") → PDF, DOCX and both reviewer reports
Notes
theses.cz is scraped, not queried through an API — it has no public one. Two consequences:
- Session bootstrap. The first request returns a 117-byte
<meta refresh>stub; the server retries once with the__Host-issessioncookie it just received. - Markup drift breaks parsing.
python theses_mcp.py --selftestruns a live check against known records and fails loudly when the selectors stop matching. Run it first when results look empty.
Be a decent citizen with request rates.
License
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.