Chroma MCP Server

Vector database access for LLM applications with collection and document management.

Data last scanned today · Reviewed today

39 2 dangerous tools12 tools★ 578PythonApache-2.0

Overview

The Chroma MCP Server enables AI models to interact directly with Chroma vector databases. It provides a structured way for assistants to manage collections, perform semantic and full-text searches, and retrieve stored documents through the Model Context Protocol. Users can connect to various Chroma client types, including ephemeral in-memory instances for testing, persistent local file storage, or remote instances via HTTP and Chroma Cloud. The server exposes specific tools for operations like creating collections, configuring HNSW parameters, adding documents with metadata, and querying data through advanced filtering.

Our verdict

This server carries a high-risk rating with a security score of 39/100. The score is heavily impacted by the fact that it is a community-maintained project rather than an official vendor release and has not undergone editorial review. Furthermore, the repository has seen no updates for 313 days, indicating a lack of active maintenance. Due to these factors, it is best suited for development or local experimentation where security oversight is managed internally, and it should be avoided in sensitive production environments until it receives more frequent maintenance and thorough review.

Supports multiple storage modes including persistent and in-memory options
Provides direct tools for collection management and advanced document querying
Contains zero direct dependencies for a smaller supply-chain surface
Includes documentation within the repository README

Security risk is high due to lack of editorial review
Project status is stale with no commits in over 300 days
Not maintained by the official vendor organization

Tools

Tool	Description	Risk
chroma_list_collections	List all collections with pagination support	low
chroma_create_collection	Create a new collection with optional HNSW configuration	medium
chroma_peek_collection	View a sample of documents in a collection	low
chroma_get_collection_info	Get detailed information about a collection	low
chroma_get_collection_count	Get the number of documents in a collection	low
chroma_modify_collection	Update a collection's name or metadata	medium
chroma_delete_collection	Delete a collection	high
chroma_add_documents	Add documents with optional metadata and custom IDs	medium
chroma_query_documents	Query documents using semantic search with advanced filtering	low
chroma_get_documents	Retrieve documents by IDs or filters with pagination	low

Compatibility

Client	Local	Docker	Remote	Read-only
Claude Desktop
Cursor
VS Code
Windsurf
ChatGPT

Frequently asked questions

›What types of Chroma clients are supported by this MCP server?

The server supports ephemeral (in-memory) clients for testing, persistent clients for file-based local storage, HTTP clients for self-hosted instances, and cloud clients for Chroma Cloud integration.

›Which embedding functions are available for use?

Supported embedding functions include default, cohere, openai, jina, voyageai, and roboflow.

›How can I securely manage API keys for external embedding functions?

While you can provide API keys directly in the configuration arguments, it is recommended to use the --dotenv-path argument to specify a custom path to your .env file containing the required environment variables.

›Are there limitations regarding older Chroma collections and embedding functions?

Yes, embedding function persistence is a feature added in Chroma v1.0.0. If your collection was created using version 0.6.3 or older, embedding function persistence is not supported.