moses-mcp
# moses-mcp
An MCP server for **discovering** degree programs and modules in [Moses](https://moseskonto.tu-berlin.de/moses/) — TU Berlin's course, exam, and module planning system. It talks to Moses' official REST API (`https://moseskonto.tu-berlin.de/moses/api/v2`), which innoCampus (who run Moses) can provision on request.
## Scope
- Search and browse degree programs (Studiengänge) and their curriculum structure (Pflichtbereich, Wahlpflichtbereich, ...).
- Search modules and read full module descriptions, including exam type (Prüfungsform) and which other degree programs use a given module.
- Out of scope: Vorlesungsverzeichnis (VVZ)/timetables, and anything requiring login (exam registration, editing module data).
## Tools
| Tool | What it does |
|---|---|
| `search_degree_programs` | Search degree programs by name, e.g. "Informatik" |
| `get_degree_program_structure` | Get a program's curriculum areas (Pflichtbereich, Wahlpflichtbereich, ...) with module/credit counts |
| `list_area_modules` | List the modules in one curriculum area, with credits, grading, exam type, and Turnus — plus the StuPO's passing rules for that area (e.g. credit min/max) |
| `search_modules` | Search modules by title or number, e.g. "Computer Vision" |
| `get_module_details` | Full module description: content, exam type, and which other degree programs use it |
Example things you can ask an MCP client connected to this server:
- "Which modules are mandatory (Pflichtmodul) for the Informatik B.Sc.?"
- "Of those, which use a Portfolioprüfung instead of a written exam?"
- "Recommend some Computer Vision modules for the Computer Science Master's."
- "What other degree programs can take module 40022?"
- "Make me a 6-semester study plan for the Informatik B.Sc. specializing in Theoretical Computer Science" (respects each area's StuPO passing rules, not just credit totals)
## Requirements
- Node.js 20 or later
- A Moses API token (see below)
## API access
There's no self-service signup — email innoCampus (who run Moses) and ask for API access; they'll issue a token and tell you which endpoints are enabled for it. Keep the token secret: it authenticates as you, and the endpoint owners can rate-limit or block API keys that generate excessive load (e.g. bulk-scanning without cause), so use `refresh: true` sparingly and avoid scripting bulk queries beyond what this tool already does.
## Setup
```bash
git clone <this-repo>
cd moses-mcp
npm install
cp .env.example .env # then edit .env and set API_TOKEN=<your token>
npm run build
```
This produces a runnable server at `dist/index.js`, which communicates over stdio (standard MCP transport). The server reads `.env` itself on startup (resolved next to its own install location, not the caller's working directory), so a `.env` file in this project directory is enough — you don't need to configure the token separately in your MCP client, though you can (see below) if you'd rather not keep it in a file.
## Using it with an MCP client
### Claude Code
```bash
claude mcp add moses -- node /absolute/path/to/moses-mcp/dist/index.js
```
This registers the server at local scope (just this project, just you). Use `-s user` to make it available in every project, or `-s project` to check a shared `.mcp.json` into this repo for teammates. Check `claude mcp list` or run `/mcp` inside a session to confirm it's connected.
### Claude Desktop / other stdio clients
Add it to your client's MCP server config, pointing at the built `dist/index.js`. For example, in Claude Desktop's `claude_desktop_config.json`:
```json
{
"mcpServers": {
"moses": {
"command": "node",
"args": ["/absolute/path/to/moses-mcp/dist/index.js"]
}
}
}
```
Any MCP client that supports stdio servers can use this the same way — point it at `node dist/index.js`. If you'd rather not keep the token in `.env`, most clients also let you set per-server environment variables directly in their config (e.g. add `"env": {"API_TOKEN": "..."}` alongside `"command"`/`"args"` above) instead.
## Enabling cross-program module usage lookups
`get_module_details`'s `usedInPrograms` (which other degree programs use a module) needs a reverse-lookup index that Moses' API doesn't support querying directly — building it means scanning the full assignment tables (~445k records), which takes on the order of 10-15 minutes. So it's never built automatically; run it explicitly, and re-run occasionally to refresh (it's cached for 7 days regardless):
```bash
npm run build-usage-index
```
Until this has been run at least once, `usedInPrograms` comes back as an empty list rather than blocking tool calls.
## Development
```bash
npm install
npm run dev # run the MCP server via stdio (tsx, no build step)
npm run build # compile to dist/
npm test # run unit tests against fixture data (no network)
npm run e2e # build, then exercise all 5 tools against the LIVE API
```
`src/api/client.ts` is the typed REST client (token auth via `X-API-Key`, batched `idlist` lookups, a small concurrency cap). `src/api/entities.ts` has the API's entity shapes. The trickiest part of `src/services/degreePrograms.ts` is that Moses' curriculum data comes in two structurally different shapes depending on the StuPO: a newer `studiengangsbereich` model, and an older per-semester `bolognamodulliste` model still used by some programs (Informatik B.Sc. among them) — `get_degree_program_structure`/`list_area_modules` detect and handle both. `src/services/moduleUsageIndex.ts` builds the cross-program usage index described above. `scripts/e2e-check.ts` runs the built server end-to-end against the live API.
## Troubleshooting
- **A degree program's curriculum looks empty / `mapped: false`**: not every Studien-/Prüfungsordnung (StuPO) has curriculum data mapped yet. `get_degree_program_structure` automatically tries a few recent StuPOs to find one with data if you don't specify `stupo`/`semester` explicitly.
- **Stale results after a Moses update**: pass `refresh: true` to any tool to bypass the cache, or delete the `.cache/` directory.
- **`get_module_details`'s `usedInPrograms` is always empty**: run `npm run build-usage-index` at least once — see above.
- **A tool errors unexpectedly**: the Moses API is under active development and has some known rough edges (e.g. `idlist` silently doesn't filter on a couple of endpoints — already worked around in `src/api/client.ts`). If a new one shows up, issues/PRs welcome.
TDQS
Scored across 5 tools
Each tool targets a distinct entity and action: searching programs, getting program structure, listing modules in an area, searching modules, and getting module details. The workflow is clear, and there is no ambiguity between search functions for different entity types.
All tool names follow a consistent verb_noun pattern in snake_case: search_degree_programs, get_degree_program_structure, list_area_modules, search_modules, get_module_details. The verbs (search, get, list) accurately reflect the operations and are used consistently.
With exactly 5 tools, the server is well-scoped for its purpose of browsing degree programs and modules. Each tool covers a necessary step in the workflow without redundancy or bloat.
The tool set covers the core workflow of discovering degree programs, exploring their curriculum areas, listing modules in those areas, and retrieving module details. A minor gap is the lack of a direct 'get_degree_program' tool by ID, though search_degree_programs returns id and basic info, and get_degree_program_structure can be used with the id.