wddxg

zotero-mcp-augmentation

Use this skill when the user wants to add local full-text cleaning, chunking, embedding, or passage retrieval to Zotero MCP. It is suitable for cleaned fulltext sidecars, chunk-level semantic search, and local RAG-style workflows built on local models instead of remote APIs.

wddxg 0 Updated 1mo ago

Resources

7
GitHub

Install

npx skillscat add wddxg/zotero-mcp-augmentation

Install via the SkillsCat registry.

SKILL.md

Zotero MCP augmentation

这个 skill 用来给 Zotero MCP 补一层本地全文处理和段落检索能力。

适用情况：

  • 需要把 OCR 或 Markdown 文本清洗后再建库
  • 需要把长文按标题结构切成可检索分段
  • 需要用本地模型生成嵌入
  • 需要按语义检索并返回段落级结果

Commands

1. Clean and chunk

python scripts/clean_and_chunk_markdown.py `
  --input source.md `
  --cleaned-out output/cleaned.md `
  --chunks-out output/chunks.jsonl `
  --title "My Document" `
  --document-id doc-001

2. Build the local store

python scripts/build_local_chunk_store.py `
  --chunks-jsonl output/chunks.jsonl `
  --out-dir output/store `
  --model BAAI/bge-m3 `
  --device cuda `
  --normalize

3. Query

python scripts/query_local_chunk_store.py `
  --store-dir output/store `
  --query "find paragraphs about coupled moisture and deformation effects"

Expected Outputs

  • cleaned.md
  • chunks.jsonl
  • rows.jsonl
  • embeddings.npy
  • manifest.json
  • query result JSON with matched documents and chunks

Output Style

执行这个 skill 时，返回内容尽量简短，优先给这些信息：

  • 输出文件路径
  • 实际执行的命令
  • 清洗时去掉了什么类型的噪声
  • 如果跑了查询，给 1 到 2 条命中示例

Files

  • scripts/clean_and_chunk_markdown.py: 清洗并分段
  • scripts/build_local_chunk_store.py: 构建本地向量库
  • scripts/query_local_chunk_store.py: 查询本地向量库
  • scripts/zotero_mcp_augmentation/: 包内实现
  • references/data-shape.md: 数据结构说明

Categories

Docs Gen Embeddings
GitHub

Install

npx skillscat add wddxg/zotero-mcp-augmentation

Install via the SkillsCat registry.

Recommended Skills