Stellaris Teacher MCP
# Stellaris Teacher MCP
Personal MCP server by Ibrahim Sabri Örene for offline-first game knowledge retrieval. Zero runtime dependencies; Node.js 18+; stdio transport.
## Capabilities
BM25-ranked knowledge search, page and section access, glossary lookup, UI-label mapping, learning paths, ingestion queues and optional timestamped video retrieval. `stellaris_explain` combines useful context for a teaching assistant.
```sh
node server.js
node test/smoke.js
```
Register `node /absolute/path/to/mcp-stellaris/server.js` with your MCP client. Set `STELLARIS_KB_DIR` to the desired writable knowledge directory. Without it, storage defaults to a sibling `.kb` directory.
## Build your own knowledge base
The public repository includes authored term mappings and source code, not a downloaded wiki or video corpus. An empty KB is expected on first run. Use `stellaris_queue_add`, `stellaris_queue_next` and `stellaris_kb_add_page` to ingest material you are permitted to use. Live HTTP access can be blocked by the wiki's browser challenge; the tool reports that limitation.
The original smoke test runs against a throwaway KB with synthetic examples. It does not use your personal corpus.
Stellaris and its wiki are third-party works. This repository is an independent tool and does not redistribute their article bodies or full video transcripts. Retrieved material retains its source rights and attribution. The project package declares MIT; see LICENSE.
TDQS
Scored across 16 tools
Most tools have clearly distinct roles (explain vs search vs glossary vs video_search), and descriptions explicitly steer between them. The main overlap is stellaris_search (which covers video too) versus stellaris_video_search, and stellaris_explain versus stellaris_search, but the descriptions differentiate them reasonably well.
All tools share the stellaris_ prefix and snake_case, which makes the set predictable. There is some mixing of verb-led (explain, search, get_page) and noun-led (glossary, outline, readme, learning_path) names, but the pattern remains readable.
16 tools is slightly on the heavy side but justified by the server's three concerns: KB querying, crawling/ingestion management, and teaching aids. Each tool earns its place, with no obvious filler.
The KB lifecycle is mostly covered: add, search, fetch, outline, related, stats, and queue management. Minor gaps exist around update/delete/re-index of stored pages, which agents can work around by re-adding content.