MCP OI-Wiki
[](https://mseep.ai/app/shwstone-mcp-oi-wiki)
[](https://mseep.ai/app/9d61dc89-76bb-401f-840e-07ba2c9cb39b)
# mcp-oi-wiki
让大模型拥有 OI-Wiki 的加成!

## How does it work?
使用 Deepseek-V3 对 OI-wiki 当前的 462 个页面做摘要,将摘要嵌入为语义向量,建立向量数据库。
查询时,找到数据库中最接近的向量,返回对应的 wiki markdown。
## Usage
确保你拥有 `uv`。
首先,下载本仓库:
```
cd <path of MCP servers>
git clone --recurse-submodules https://github.com/ShwStone/mcp-oi-wiki.git
```
然后打开你的 MCP 配置文件(mcpo 或 claude):
```json
{
"mcpServers": {
"oi-wiki": {
"command": "uv",
"args": [
"--directory",
"<path of MCP servers>/mcp-oi-wiki",
"run",
"python",
"main.py"
]
}
}
}
```
## Update
可以生成自己的 `db/oi-wiki.db`。
将 Silicon flow API key 放在 `api.key` 文件中。
然后运行:
```sh
uv run script/request.py
```
在[批量推理页面](https://cloud.siliconflow.cn/batches)下载摘要结果到 `result.jsonl`。
最后运行:
```sh
uv run script/gendb.py
```
生成新的 `db/oi-wiki.db`。
## Thanks
- [milvus-io/milvus-lite: A lightweight version of Milvus](https://github.com/milvus-io/milvus-lite) 向量数据库
- [OI-wiki/OI-wiki: :star2: Wiki of OI / ICPC for everyone. (某大型游戏线上攻略,内含炫酷算术魔法)](https://github.com/OI-wiki/OI-wiki) OI-wiki
- [qdrant/fastembed: Fast, Accurate, Lightweight Python library to make State of the Art Embedding](https://github.com/qdrant/fastembed) CPU 向量嵌入
TDQS
Scored across 1 tool
With only one tool, there is no possibility of ambiguity or overlap between tools. The single tool has a clearly defined and distinct purpose: searching the OI Wiki for programming competition knowledge.
The single tool name 'search' follows a simple, clear verb pattern. Since there is only one tool, consistency is inherently perfect with no deviations or mixed conventions to evaluate.
A single tool is too few for a server that appears to cover a broad domain like a programming competition knowledge base. The scope suggests needs for browsing, filtering, or accessing specific content beyond just search, making the toolset feel thin and underdeveloped.
The tool surface is severely incomplete for the domain. While search is useful, there are obvious gaps such as listing topics, getting detailed articles, or navigating categories, which limits an agent's ability to fully interact with the wiki's content.