mcp-oi-wiki

Empower large models with the OI-Wiki boost!

How does it work?

We use Deepseek-V3 to generate summaries of the current 462 pages on OI-wiki, embed these summaries as semantic vectors, and establish a vector database.

During queries, the closest vector in the database is found, and the corresponding wiki markdown is returned.

Usage

Ensure you have uv.

First, download this repository:

cd <path of MCP servers> git clone --recurse-submodules https://github.com/ShwStone/mcp-oi-wiki.git

Then, open your MCP configuration file (mcpo or claude):

json { "mcpServers": { "oi-wiki": { "command": "uv", "args": [ "--directory", "<path of MCP servers>/mcp-oi-wiki", "run", "python", "main.py" ] } } }

Update

You can generate your own db/oi-wiki.db.

Place your Silicon flow API key in the api.key file.

Then run:

sh uv run script/request.py

Download the summary results to result.jsonl from the batch inference page.

Finally, run:

sh uv run script/gendb.py

This will generate a new db/oi-wiki.db.

Thanks

• milvus-io/milvus-lite: A lightweight version of Milvus for the vector database
• OI-wiki/OI-wiki: :star2: Wiki of OI / ICPC for everyone. (A comprehensive online guide for a popular game, featuring cool arithmetic magic) for OI-wiki
• qdrant/fastembed: Fast, Accurate, Lightweight Python library to make State of the Art Embedding for CPU-based vector embedding