theses-mcp
# theses-mcp
Search Czech university theses and retrieve document links for MCP clients.
[](https://github.com/koprjaa/theses-mcp/actions/workflows/ci.yml)
[](LICENSE)
## What it does
- Searches thesis metadata and full texts on theses.cz.
- Reads titles, abstracts, keywords, and defense information.
- Finds thesis files, attachments, and reviewer reports on theses.cz or school repositories.
- Supports your own authenticated sessions through `login`, `use_cookie`, and `whoami`.
## Install
Install a current stable Rust toolchain, then:
```bash
cargo install --git https://github.com/koprjaa/theses-mcp
```
This puts `theses-mcp` in `~/.cargo/bin`. Add this entry to your MCP client's `mcpServers` configuration:
```json
{
"theses": {
"command": "theses-mcp"
}
}
```
No environment variables are required for public access.
## Use
Ask your MCP client to call these tools:
```text
search(query="midjourney AND autorství", limit=10)
detail(id_or_url="pl09jx")
fulltext(id_or_url="7lfo74")
```
Use a record's code or URL for `detail` and `fulltext`. Search accepts `AND`, `OR`, and quoted phrases. Your client exposes the remaining tool parameters.
`fulltext` returns file links, not extracted text. Check its `files` and `note` fields for availability. Some repositories require login or show a CAPTCHA. Call `login` with the repository host to sign in using your own account. Sessions persist in `~/.theses-mcp/cookies.json`; delete that file to forget them. The file format is the one the earlier Python version wrote, so saved sessions keep working.
## Browser login
`login` opens a window of a Chromium-family browser (Chrome, Edge, Brave, Vivaldi, Opera, or Chromium) with a profile of its own in `~/.theses-mcp/browser`, and reads back only the cookies of the host you sign in to. It talks to the browser over the Chrome DevTools Protocol, so nothing is downloaded.
- On Windows it opens your default browser if that is a Chromium build. On macOS and Linux there is no cheap way to ask, so it uses the first Chromium-family browser it finds at the usual places.
- Firefox and Safari cannot be driven. Use `use_cookie` with a cookie copied from the browser's developer tools instead.
- `attach=true` uses a browser that is already running with `--remote-debugging-port`.
## Check
`theses-mcp --selftest` runs the parsers against the live sites. It makes several dozen requests, so run it by hand now and then, not in CI.
## License
[MIT](LICENSE)
TDQS
Scored across 3 tools
Each tool targets a distinct stage of working with theses.cz: search finds theses or fulltext snippets, detail retrieves full metadata for a single record, and fulltext resolves a thesis to downloadable files in its school repository. Even though search and fulltext both deal with PDFs, their purposes are clearly separate.
All three tool names are single lowercase words (search, detail, fulltext), following a consistent, minimalist naming convention. While not verb_noun, the style is uniform and predictable, making the set easy to navigate.
With only three tools, the server is tightly scoped for its purpose: searching, inspecting, and retrieving full text from a national thesis registry. Each tool earns its place, and the count fits well within the typical 3-15 range for a focused domain.
The toolset covers the complete user journey for the domain: discover via search, examine metadata via detail, and access actual files via fulltext. There are no obvious gaps—it's a read-only service, and lifecycle operations like create/update are not applicable.