<img height="24" width="24" src="https://docs.trychroma.com/favicon.ico" alt="ChromaDB Logo" /> Chroma MCP Server

A Model Context Protocol (MCP) server that provides semantic search and document management capabilities using ChromaDB. This server enables LLMs to perform natural language queries over document collections with intuitive similarity metrics, making it ideal for RAG (Retrieval Augmented Generation) applications.

Features

Semantic Search: Find documents based on meaning using state-of-the-art embeddings
Intuitive Similarity Metrics: Results include human-friendly similarity scores (0-100%)
Document Management: Full CRUD operations for documents and collections
Rich Metadata Support: Attach and search by custom metadata fields
Persistent Storage: Reliable document storage with SQLite backend
Security: Configurable access controls and input validation
Error Handling: Comprehensive error messages and graceful failure recovery

Requirements

Python 3.12 or higher
ChromaDB 0.4.22 or higher
MCP Python SDK 1.1.2 or higher
uv package manager (recommended) or pip

Quick Start

# Clone the repository
git clone https://github.com/privetin/mcp-server-chroma.git
cd mcp-server-chroma

# Install with uv (recommended)
uv venv
.venv\Scripts\activate  # Windows
source .venv/bin/activate  # Unix
uv pip install -e .

# Or with pip
python -m venv .venv
.venv\Scripts\activate  # Windows
source .venv/bin/activate  # Unix
pip install -e .

# Run the server
mcp-server-chroma

For Claude Desktop integration, see Installation.

Architecture

The server is built on:

ChromaDB for vector storage and search
MCP Python SDK for server implementation
SQLite for persistent storage

Data Flow

Documents are embedded using ChromaDB's default embedding model
Embeddings and metadata are stored in ChromaDB's SQLite backend
Queries are processed through the same embedding model
Results are normalized to a 0-100% similarity scale

Components

Collections and Documents

The server manages two main resource types:

Collections: Containers for related documents with shared embedding settings
Documents: Text content with metadata and automatically generated embeddings

Tools

Collection Management

list-collections: List all available collections
create-collection: Create a new collection with optional settings
delete-collection: Delete a collection and its documents

Document Operations

add-document: Add a new document with content and metadata
get-document: Retrieve a specific document by ID
update-document: Modify document content or metadata
delete-document: Remove a document from a collection
search-documents: Semantic search with normalized similarity scores

Installation

Prerequisites

Python 3.12+
uv package manager (recommended) or pip

Setup

Clone the repository:

git clone https://github.com/privetin/mcp-server-chroma.git
cd mcp-server-chroma

Create and activate virtual environment:

uv venv
# On Windows:
.venv\Scripts\activate
# On Unix:
source .venv/bin/activate

Install dependencies:

uv pip install -e .

Claude Desktop Integration

Add the server to your Claude Desktop configuration:

Windows (%APPDATA%/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "chroma": {
      "command": "uv",
      "args": [
        "--directory",
        "C:\\path\\to\\mcp-server-chroma",
        "run",
        "mcp-server-chroma"
      ]
    }
  }
}

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

{
  "mcpServers": {
    "chroma": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/mcp-server-chroma",
        "run",
        "mcp-server-chroma"
      ]
    }
  }
}

Usage Examples

Managing Collections

Create a collection:

Tool: create-collection
Args: {"name": "research-papers"}

List collections:

Tool: list-collections
Args: {}

Working with Documents

Add a document:

Tool: add-document
Args: {
  "collection": "research-papers",
  "content": "Recent advances in transformer architectures have led to significant improvements in natural language processing tasks.",
  "metadata": {
    "title": "Transformer Architectures",
    "year": 2024,
    "category": "ML"
  }
}

Get a specific document:

Tool: get-document
Args: {
  "collection": "research-papers",
  "document_id": "doc_123"
}

Update a document:

Tool: update-document
Args: {
  "collection": "research-papers",
  "document_id": "doc_123",
  "content": "Updated findings on transformer architectures show improvements in both efficiency and accuracy.",
  "metadata": {
    "title": "Transformer Architectures - Updated",
    "year": 2024,
    "category": "ML",
    "status": "updated"
  }
}

Search documents:

Tool: search-documents
Args: {
  "collection": "research-papers",
  "query": "What are the latest developments in transformers?",
  "n_results": 3
}

Understanding Similarity Scores

Search results include normalized similarity scores from 0-100%:

90-100%: Nearly identical content or very strong semantic match
70-89%: Highly relevant with strong semantic similarity
50-69%: Moderately related with partial semantic overlap
30-49%: Somewhat related with minimal semantic connection
0-29%: Likely unrelated or very weak semantic connection

Troubleshooting

Common Issues

Database Connection Errors
- Ensure the database path is writable
- Check if another process is using the database
- Try deleting .chroma directory and restarting
Memory Issues
- Large collections may require more RAM
- Consider using smaller batch sizes
- Monitor memory usage with --log-level DEBUG
Slow Search Performance
- Large collections may need index optimization
- Consider using fewer n_results
- Check system resource usage

Debug Mode

Run the server in debug mode:

mcp-server-chroma --log-level DEBUG

Getting Help

Check ChromaDB Documentation
Open an issue on GitHub
Join MCP Community Discussions

Development

Running Tests

Run the test suite:

pytest -v

Run with coverage:

pytest --cov=chroma tests/

Debugging

For debugging, use the MCP Inspector:

# Install the inspector
npm install -g @modelcontextprotocol/inspector

# Run the server with inspector
mcp-inspector uv --directory /path/to/mcp-server-chroma run mcp-server-chroma

The inspector provides:

Real-time request/response monitoring
Tool testing interface
Performance metrics
Error tracking

Error Handling

The server provides detailed error messages for common scenarios:

Invalid collection names or IDs
Missing or malformed documents
Database connection issues
Invalid search parameters
Authentication/authorization failures

Security Considerations

Input validation on all parameters
Configurable access controls
Safe handling of file paths
Protection against injection attacks
Rate limiting support
Secure error messages

Configuration

Database Location

Set custom database path:

mcp-server-chroma --db-path /path/to/db

Default: .chroma in the server directory

Environment Variables

CHROMA_DB_PATH: Override database location
CHROMA_LOG_LEVEL: Set logging verbosity (default: INFO)
CHROMA_MAX_CONNECTIONS: Database connection pool size (default: 10)

Contributing

Fork the repository
Create a feature branch
Make your changes
Add tests for new functionality
Submit a pull request

Please read our Contributing Guidelines for more details.

License

MIT License

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.