Chroma

Chroma MCP Server

A Model Context Protocol (MCP) server implementation that provides vector database capabilities through Chroma. This server enables semantic document search, metadata filtering, and document management with persistent storage.

Requirements

• Python 3.8+
• Chroma 0.4.0+
• MCP SDK 0.1.0+

Components

Resources

The server provides document storage and retrieval through Chroma's vector database:

• Stores documents with content and metadata
• Persists data in src/chroma/data directory
• Supports semantic similarity search

Tools

The server implements CRUD operations and search functionality:

Document Management

• create_document: Create a new document

  • Required: document_id, content
  • Optional: metadata (key-value pairs)
  • Returns: Success confirmation
  • Error: Already exists, Invalid input

• read_document: Retrieve a document by ID

  • Required: document_id
  • Returns: Document content and metadata
  • Error: Not found

• update_document: Update an existing document

  • Required: document_id, content
  • Optional: metadata
  • Returns: Success confirmation
  • Error: Not found, Invalid input

• delete_document: Remove a document

  • Required: document_id
  • Returns: Success confirmation
  • Error: Not found

• list_documents: List all documents

  • Optional: limit, offset
  • Returns: List of documents with content and metadata

Search Operations

• search_similar: Find semantically similar documents
  • Required: query
  • Optional: num_results, metadata_filter, content_filter
  • Returns: Ranked list of similar documents with distance scores
  • Error: Invalid filter

Features

• Semantic Search: Find documents based on meaning using Chroma's embeddings
• Metadata Filtering: Filter search results by metadata fields
• Content Filtering: Additional filtering based on document content
• Persistent Storage: Data persists in local directory between server restarts
• Error Handling: Comprehensive error handling with clear messages
• Retry Logic: Automatic retries for transient failures

Installation

1. Install dependencies:

1uv venv
2uv sync --dev --all-extras

Configuration

Claude Desktop

Add the server configuration to your Claude Desktop config:

Windows: C:\Users\<username>\AppData\Roaming\Claude\claude_desktop_config.json

MacOS: ~/Library/Application Support/Claude/claude_desktop_config.json

1{
2  "mcpServers": {
3    "chroma": {
4      "command": "uv",
5      "args": [
6        "--directory",
7        "C:/MCP/server/community/chroma",
8        "run",
9        "chroma"
10      ]
11    }
12  }
13}

Data Storage

The server stores data in:

• Windows: src/chroma/data
• MacOS/Linux: src/chroma/data

Usage

1. Start the server:

1uv run chroma

2. Use MCP tools to interact with the server:

1# Create a document
2create_document({
3    "document_id": "ml_paper1",
4    "content": "Convolutional neural networks improve image recognition accuracy.",
5    "metadata": {
6        "year": 2020,
7        "field": "computer vision",
8        "complexity": "advanced"
9    }
10})
11
12# Search similar documents
13search_similar({
14    "query": "machine learning models",
15    "num_results": 2,
16    "metadata_filter": {
17        "year": 2020,
18        "field": "computer vision"
19    }
20})

Error Handling

The server provides clear error messages for common scenarios:

• Document already exists [id=X]
• Document not found [id=X]
• Invalid input: Missing document_id or content
• Invalid filter
• Operation failed: [details]

Development

Testing

1. Run the MCP Inspector for interactive testing:

1npx @modelcontextprotocol/inspector uv --directory C:/MCP/server/community/chroma run chroma

2. Use the inspector's web interface to:
   • Test CRUD operations
   • Verify search functionality
   • Check error handling
   • Monitor server logs

Building

1. Update dependencies:

1uv compile pyproject.toml

2. Build package:

1uv build

Contributing

Contributions are welcome! Please read our Contributing Guidelines for details on:

• Code style
• Testing requirements
• Pull request process

License

This project is licensed under the MIT License - see the LICENSE file for details.