mcp-oi-wiki

What is mcp-oi-wiki?

mcp-oi-wiki is an MCP server that provides large language models with access to OI-Wiki content. It uses Deepseek-V3 to summarize 462 OI-Wiki pages, embeds the summaries as semantic vectors, and returns the matching wiki markdown for nearest vectors on query.

How to use mcp-oi-wiki?

Ensure you have uv. Clone the repository with git clone --recurse-submodules https://github.com/ShwStone/mcp-oi-wiki.git, then add the following to your MCP configuration file (for mcpo or Claude):

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

Key features of mcp-oi-wiki

• Summarizes 462 OI-Wiki pages via Deepseek-V3.
• Semantic vector search using Milvus Lite.
• CPU embedding generation via fastembed.
• Lightweight, self-contained MCP server.
• Provides raw wiki markdown for query matches.

Use cases of mcp-oi-wiki

• Quick retrieval of OI-Wiki content through AI assistants.
• Adding competitive programming knowledge to LLM workflows.
• Offline-capable search of OI-Wiki summaries.
• Enriching AI responses with verified OI/ICPC reference material.

FAQ from mcp-oi-wiki

What does mcp-oi-wiki do?

It gives large language models “OI-Wiki 的加成” (the advantage of OI-Wiki) by turning wiki pages into a searchable vector database.

How does the vector search work?

OI-Wiki pages are summarized using Deepseek-V3, embedded as semantic vectors, and stored in a Milvus Lite database. On query, the closest vector is found and its corresponding wiki markdown is returned.

Can I update the database myself?

Yes. You can generate your own db/oi-wiki.db by placing a Silicon Flow API key in api.key, running uv run script/request.py, downloading the batch inference results from the Silicon Flow platform, and finally running uv run script/gendb.py.

What platforms or editors does it support?

The README explicitly mentions it works with MCP-compatible clients such as mcpo and Claude. No other platforms are listed.

Are there any limits or licensing details?

The README does not mention pricing, licensing, or limits beyond that it currently uses 462 OI-Wiki pages.