Best for
- Vector database scenarios that need integration with MCP-compatible clients like Claude.
- Developers and teams who want to self-host Chroma and unify APIs.
- AI applications that require persistent or cloud-hosted vector storage.
C
Chroma MCP Server
Official
Model Context Protocol server for Chroma, providing database capabilities for AI models.
GitHub source repository ↗
48FMRS · D
This server is officially maintained by the Chroma team, feature-complete, and easy to integrate. It is suitable for developers who need to add vector memory to AI applications, but be aware of external API and data security risks.
This is an MCP server implementation for Chroma, an open-source embedding database. It allows AI models (e.g., LLMs) to interact with Chroma via the standardized Model Context Protocol, enabling data storage, semantic search, and retrieval. It supports ephemeral (in-memory), persistent, HTTP self-hosted, and cloud (Chroma Cloud) client types, and offers collection management, document operations, vector search, full-text search, and more.
Tools
chroma_list_collections
List all collections with pagination support.
chroma_create_collection
Create a new collection with optional HNSW configuration.
chroma_peek_collection
View a sample of documents in a collection.
chroma_get_collection_info
Get detailed information about a collection.
chroma_get_collection_count
Get the number of documents in a collection.
chroma_modify_collection
Update a collection's name or metadata.
chroma_delete_collection
Delete a collection.
chroma_add_documents
Add documents with optional metadata and custom IDs.
chroma_query_documents
Query documents using semantic search with advanced filtering.
chroma_get_documents
Retrieve documents by IDs or filters with pagination.
chroma_update_documents
Update existing documents' content, metadata, or embeddings.
chroma_delete_documents
Delete specific documents from a collection.
Setup
- Install Python and uv (or pip).
- Add the MCP server configuration to your Claude Desktop config file, using uvx to run chroma-mcp.
- Configure client type as needed (default ephemeral; can switch to persistent, HTTP, or cloud via arguments).
claude_desktop_config.json
{
"mcpServers": {
"chroma": {
"command": "uvx",
"args": [
"chroma-mcp"
]
}
}
}
Fit and risk
Not for
- Not suitable for non-vector database use cases.
- Not for applications needing relational queries or complex transactions.
- Not ideal for scenarios with high data privacy requirements that can't tolerate external API calls (when using external embedding functions).
Required permissions
- File system access (for persistent client) to store data.
- Network access (for HTTP or cloud clients) to connect to remote Chroma instances or external embedding APIs.
- API keys in environment variables or config files (e.g., Cohere, OpenAI) to call external embedding models.
Risks and side effects
- External embedding APIs may send data to third-party services, posing data leakage risks.
- Self-hosted HTTP clients without proper authentication may expose sensitive data.
- Persistent clients require write permissions and backup for the data directory.
Troubleshooting
- Ensure uv is installed or the Python environment is set up correctly.
- Check that .env file or environment variables are correctly set, especially API keys.
- For HTTP clients, confirm host, port, and SSL settings are correct and the service is reachable.
- For persistent clients, check that the data directory exists and is writable.
Use cases
Add long-term memory to LLM applications by storing and retrieving conversation history or knowledge bases.
Build semantic search features to retrieve documents based on vector similarity.
Quickly test vector database functionality in development using in-memory mode without extra deployment.
Supported clients
Claude DesktopFull support