语义搜索工具

{
  "mcpServers": {
    "context-lens": {
      "command": "uvx",
      "args": ["context-lens"]
    }
  }
}

claude mcp add context-lens -- npx context-lens@latest

{
  "mcpServers": {
    "context-lens": {
      "command": "uvx",
      "args": ["context-lens"]
    }
  }
}

{
  "mcpServers": {
    "context-lens": {
      "command": "uvx",
      "args": ["context-lens"],
      "disabled": false,
      "autoApprove": ["list_documents", "search_documents"]
    }
  }
}

{
  "mcpServers": [
    {
      "name": "context-lens",
      "command": "uvx",
      "args": ["context-lens"]
    }
  ]
}

codex mcp add context-lens -- npx @context-lens@latest

{
  "mcpServers": {
    "context-lens": {
      "command": "node",
      "args": ["/absolute/path/to/context-lens/dist/index.js"]
    }
  }
}

{
  "command": "uvx",
  "args": ["context-lens"]
}

{
  "mcpServers": {
    "context-lens": {
      "command": "uvx",
      "args": ["context-lens"],
      "env": {
        "LANCE_DB_PATH": "./my_knowledge_base.db"
      }
    }
  }
}

pip install context-lens

git clone https://github.com/cornelcroi/context-lens.git
cd context-lens
pip install -e .

https://github.com/django/django          # Django web framework
https://github.com/pallets/flask          # Flask microframework  
https://github.com/fastapi/fastapi        # FastAPI modern framework

https://github.com/pandas-dev/pandas      # Pandas data analysis
https://github.com/scikit-learn/scikit-learn  # Machine learning
https://github.com/pytorch/pytorch        # PyTorch deep learning

https://github.com/psf/requests           # HTTP library
https://github.com/python/cpython         # Python itself!
https://github.com/pallets/click          # CLI framework

https://github.com/django/django/tree/main/django/contrib/auth  # Django auth
https://github.com/fastapi/fastapi/tree/master/fastapi          # FastAPI core
https://github.com/requests/requests/tree/main/requests         # Requests lib

You: Add the FastAPI repository to my knowledge base

LLM: I'll add the FastAPI repository for you.
     [Calls add_document("https://github.com/tiangolo/fastapi")]
     
     Cloning repository...
     Processing 247 Python files...
     ✓ Added 247 files from repository with 1,543 chunks
     
     You can now ask questions about FastAPI's implementation!

You: How does FastAPI handle dependency injection?

LLM: [Searches the FastAPI codebase...]
     Based on the FastAPI source code, dependency injection works through...

You: Add all Python files in my src/ directory to the knowledge base

LLM: I'll add those files for you.
     [Calls add_document for each .py file]
     ✓ Added 15 Python files to the knowledge base

You: How do we handle authentication in this codebase?

LLM: Let me search for authentication-related code.
     [Calls search_documents with query "authentication handling"]
     
     Based on the code, you use JWT tokens with OAuth2. Here's what I found:
     - src/auth/jwt.py: Token generation and validation
     - src/auth/oauth.py: OAuth2 flow implementation
     - src/middleware/auth.py: Authentication middleware
     
     The main authentication flow is...

You: Add all Python files from /Users/me/my-project/src to the knowledge base

LLM: [Processes all .py files in the directory]
     ✓ Added 23 Python files with 156 chunks

You: Add the FastAPI repository so I can learn how it works

LLM: [Calls add_document("https://github.com/tiangolo/fastapi")]
     Cloning repository...
     ✓ Added 247 files from repository with 1,543 chunks
     
You: How does FastAPI handle dependency injection?

LLM: [Searches and explains based on actual FastAPI source code]

You: Add just the authentication module from Django

LLM: [Calls add_document("https://github.com/django/django/tree/main/django/contrib/auth")]
     ✓ Added 45 files from django/contrib/auth with 312 chunks

You: Show me how Django implements password hashing

LLM: [Provides detailed explanation with code references]

"How does this codebase handle database connections?"
"Explain the authentication flow in this project"
"What design patterns are used in this repository?"
"How is error handling implemented?"
"Show me how the API endpoints are structured"

"Find examples of how to handle file uploads"
"Where is the rate limiting logic implemented?"
"Show me similar error handling patterns"
"How do other files handle this exception?"
"Find all places where we validate user input"

"How does FastAPI implement dependency injection?"
"Compare how Django and Flask handle routing"
"What's the difference between these two implementations?"
"Show me examples of async/await usage in this codebase"
"How does this library handle backwards compatibility?"

"Find all files that use the old authentication method"
"Where else do we use this deprecated function?"
"Show me similar code that might have the same bug"
"Find duplicate logic that could be refactored"
"What files would be affected if I change this interface?"

"How do I use the caching system in this project?"
"Show me examples of writing tests for API endpoints"
"How is configuration managed in this codebase?"
"Find examples of custom middleware implementation"
"How do I add a new database model?"

"How does React implement hooks internally?"
"Show me how Django's ORM builds SQL queries"
"How does FastAPI achieve such high performance?"
"Explain how pytest's fixture system works"
"How does requests handle HTTP retries?"

{
  "mcpServers": {
    "context-lens": {
      "command": "context-lens"
    }
  }
}

{
  "mcpServers": {
    "context-lens": {
      "command": "python",
      "args": ["-m", "context_lens.main"],
      "disabled": false,
      "autoApprove": ["list_documents", "search_documents"]
    }
  }
}

# Test with MCP Inspector
DANGEROUSLY_OMIT_AUTH=true npx @modelcontextprotocol/inspector python -m context_lens.server

1. 📄 Document Ingestion
   ├─ Read file content with encoding detection
   ├─ Generate content hash for deduplication
   ├─ Extract metadata (size, timestamps, type)
   └─ Split into overlapping chunks (~1000 chars)

2. 🧮 Vector Embedding
   ├─ Load sentence-transformers model (all-MiniLM-L6-v2)
   ├─ Convert each chunk to 384-dimensional vector
   ├─ Batch processing for efficiency
   └─ Store vectors in LanceDB

3. 🔍 Semantic Search
   ├─ Convert search query to vector
   ├─ Find similar vectors using ANN (Approximate Nearest Neighbor)
   ├─ Rank results by cosine similarity
   └─ Return relevant chunks with metadata

4. 💾 Storage
   ├─ LanceDB: Fast columnar vector database
   ├─ Two tables: documents + chunks
   ├─ Efficient updates and deletes
   └─ All data stays local

You: "Find authentication code"
Result: Files containing the word "authentication"
Problem: Misses related concepts like "login", "auth", "credentials"

You: "Find authentication code"  
Result: All auth-related code including:
  ✓ Files about "login" and "sign in"
  ✓ Code handling "credentials" and "tokens"
  ✓ "Authorization" and "access control"
  ✓ Related security implementations

Why: Understands meaning, not just words

# Check installation
context-lens --version

# View detailed logs
tail -f logs/context-lens.log

# Check for errors
tail -f logs/errors.log

# Make sure you're using the correct command
npx @modelcontextprotocol/inspector python -m context_lens.server

# NOT this (incorrect):
# npx @modelcontextprotocol/inspector fastmcp run context_lens.server:app

# Check that Python can find the module
python -m context_lens.server --help

# Check database location
ls -la knowledge_base.db

# If corrupted, you can reset it
rm -rf knowledge_base.db
# The server will create a new database on next run

# Reinstall dependencies
pip install -r requirements.txt

# For development installation
pip install -e .

# Check environment variables
env | grep MCP_KB

# Use config file for complex setups
cp config.example.yaml config.yaml
# Edit config.yaml with your settings
context-lens --config config.yaml

git clone https://github.com/yourusername/context-lens.git
cd context-lens
pip install -e .
pytest tests/

{
  "env": {
    "LANCE_DB_PATH": "./context-lens.db",
    "LOG_LEVEL": "INFO"
  }
}