cowork-semantic-search

Local semantic search over documents (txt, md, pdf, docx, pptx, csv). Fully offline, multilingual, hybrid vector + keyword search via LanceDB. No API keys, no cloud.

If you find this useful, consider giving it a ⭐ — it helps others discover the project.

Local semantic search for your documents. No API keys. No cloud. Works with any MCP client.

AI coding tools are powerful, but they have blind spots when it comes to your local files:

- Frozen knowledge-- training data has a cutoff. Your latest reports, notes, and contracts don't exist in the model's world.
- Context window limits-- you can't paste 500 documents into a prompt.
- No cross-file search-- your AI tool can read one file at a time, but can't search across your entire document library for the relevant pieces.

This plugin bridges that gap. It indexes your local documents into a small, fast vector database. When you ask a question, it retrieves only the relevant pieces -- so your AI tool can answer with your actual data.

Your documents --> chunked --> embedded --> local vector DB | Your question --> embedded --> similarity search --> relevant chunks --> AI answers

- Fully offline-- one-time model download (~120MB), then no network calls. No data leaves your machine.
- Incremental indexing-- SHA-256 content hashing. Only changed files get reprocessed. Re-indexing 1000 files where 3 changed takes seconds.
- Multilingual-- handles 50+ languages natively. Search in one language, find results in another.
- Hybrid search-- combines semantic similarity with full-text keyword search via Reciprocal Rank Fusion. Catches what pure vector search misses.
- Multiple formats-- txt, md, pdf, docx, pptx, csv out of the box.
- Any MCP client-- works with Claude Code, Cursor, Windsurf, Cline, and any other MCP-compatible tool.
- Zero infrastructure-- LanceDB stores everything as local files. No server, no Docker, no database to manage.

git clone https://github.com/ZhuBit/cowork-semantic-search.git cd cowork-semantic-search python3 -m venv .venv && source .venv/bin/activate pip install -e ".[all]"

Add the server to your MCP client's config. Replace paths with your own.

{ "mcpServers": { "semantic-search": { "command": "/absolute/path/to/.venv/bin/python", "args": ["-m", "server.main"], "cwd": "/absolute/path/to/cowork-semantic-search", "env": { "PYTHONPATH": "/absolute/path/to/cowork-semantic-search" } } } }

{ "mcpServers": { "semantic-search": { "command": "/absolute/path/to/.venv/bin/python", "args": ["-m", "server.main"], "env": { "PYTHONPATH": "/absolute/path/to/cowork-semantic-search" } } } }

{ "mcpServers": { "semantic-search": { "command": "/absolute/path/to/.venv/bin/python", "args": ["-m", "server.main"], "env": { "PYTHONPATH": "/absolute/path/to/cowork-semantic-search" } } } }

Open Cline > MCP Servers icon > Configure > Advanced MCP Settings, then add:

{ "mcpServers": { "semantic-search": { "command": "/absolute/path/to/.venv/bin/python", "args": ["-m", "server.main"], "env": { "PYTHONPATH": "/absolute/path/to/cowork-semantic-search" } } } }

"Index all documents in ~/Documents/projects"

"Search for 'quarterly revenue report'"

First run downloads the embedding model (~120MB), then everything runs offline.

If you keep notes in Obsidian (or any folder of markdown files), this plugin turns your AI tool into a search engine for your knowledge base.

You: "Index my vault at ~/Documents/ObsidianVault" AI: Indexed 847 files -> 3,291 chunks in 42s You: "What did I write about API rate limiting?" AI: Found 6 relevant chunks across 3 files: - notes/backend/rate-limiting-strategies.md - projects/acme-api/design-decisions.md - daily/2025-11-03.md ... You: "Find anything about the client meeting last November, use hybrid search" AI: Found 4 results using hybrid search (vector + keyword): - meetings/2025-11-12-acme-kickoff.md - daily/2025-11-12.md ...

Works the same with PDFs, Word docs, PowerPoints, and CSVs -- just point it at a folder.
- Parse-- extract text from each document, preserving structure (pages, slides)
- Chunk-- split into ~400 character overlapping pieces for precise retrieval
- Embed-- convert each chunk into a 384-dimensional vector usingparaphrase-multilingual-MiniLM-L12-v2
- Store-- save chunks + vectors in a LanceDB database (a local file, no server needed)
- Search-- embed your query, find nearest chunks by cosine similarity, optionally combine with full-text keyword search via RRF

from server.indexer import index_folder from server.search import semantic_search # Index a folder result = index_folder("/path/to/docs") print(f"{result['files_indexed']} files -> {result['total_chunks']} chunks") # Search results = semantic_search("project deadline", mode="hybrid") for r in results["results"]: print(f" {r['file_name']}: {r['text'][:100]}...")

server/ main.py # MCP server + tool definitions parsers.py # Per-format text extraction chunker.py # Text splitting with metadata indexer.py # Discovery, hashing, embedding pipeline store.py # LanceDB vector store + FTS + hybrid search search.py # Query embedding + search orchestration

source .venv/bin/activate pytest tests/ -v

56 tests covering parsers, chunking, indexing, search, and MCP tool integration.

Contributions welcome -- open an issue or submit a PR.

- ONNX runtime for faster embeddings (drop PyTorch dependency)
- Configurable chunk size and overlap via tool params
- Multi-folder named indexes
- Metadata filtering (date ranges, tags, custom fields)
- Watch mode (auto-reindex on file changes)

If this is useful to you, consider giving it a ⭐ — it helps others find the project.

AGPL-3.0 -- free to use, modify, and self-host. If you offer this as a network service, you must share your source code. SeeLICENSEfor details.

MCP server for Apache AGE graph databases on PostgreSQL. 21 MCP tools — the most comprehensive Apache AGE MCP server (graph CRUD, Cypher queries, batch transactions, semantic search, Graph RAG, vis.js visualization, export/import) - F#/.NET — the only non-Python Apache AGE MCP server, installs as a single dotnet tool - Production-grade — BenchmarkDotNet-verified performance (cached queries in 62 ns, Cypher in 1 ms) - Open source — MIT license, published on NuGet - Documentation — full docs site at neftedollar.com/age-mcp

An MCP server for vector storage and retrieval using ChromaDB.

Persistent AI memory server with 3-layer hybrid search, confidence scoring, and 16 tools. Zero LLM dependency.

Query and interact with FalkorDB graph databases using AI models.

A free, plug-and-play knowledge base with over 10,000 built-in insight reports and support for parsing private documents.

Interact with on-disk documents using agentic RAG and hybrid search via LanceDB.

A vector database server for storing, searching, and managing vector embeddings.