Chroma MCP Server

Vector database MCP server from Chroma for collection management, document operations, and semantic search over knowledge bases.

Data & Enrichment by Chroma API Key active

GitHub Docs

Overview

Chroma MCP Server is the official Model Context Protocol integration for Chroma, an open-source vector database. It lets LLM agents create and manage collections, ingest documents with metadata, and perform vector search, full-text search, and metadata filtering, making it a practical building block for retrieval-augmented generation and persistent agent memory.

The server exposes 12 tools covering the full document lifecycle: collection CRUD, document insertion, semantic queries, and updates/deletes. It supports four client modes: ephemeral (in-memory for development), persistent (local file-based storage), HTTP (self-hosted Chroma instances), and Cloud (managed Chroma Cloud). Embedding functions can be swapped between the default model, OpenAI, Cohere, Jina, VoyageAI, and Roboflow.

The repo is maintained by chroma-core (the Chroma team) and is distributed as a Python package runnable via uvx chroma-mcp. It is licensed under Apache 2.0 and is the recommended way to give Claude or other MCP-aware clients durable, queryable knowledge stores.

Tools

Tool	Description
`chroma_list_collections`	List collections with pagination support.
`chroma_create_collection`	Create a new collection with optional HNSW configuration and embedding function.
`chroma_peek_collection`	Sample a small number of documents from a collection for inspection.
`chroma_get_collection_info`	Retrieve metadata and configuration details for a collection.
`chroma_get_collection_count`	Return the total number of documents in a collection.
`chroma_modify_collection`	Update an existing collection's name or metadata.
`chroma_delete_collection`	Permanently delete a collection.
`chroma_add_documents`	Insert documents with optional metadata and custom IDs.
`chroma_query_documents`	Semantic search across documents with advanced metadata filtering.
`chroma_get_documents`	Retrieve documents by ID or filter with pagination.
`chroma_update_documents`	Modify existing documents' content, metadata, or embeddings.
`chroma_delete_documents`	Delete specific documents from a collection by ID.

Setup Guide

Prerequisites

uv / uvx installed (Python package runner)
For Cloud: a Chroma Cloud account with tenant ID, database name, and API key
For HTTP: a running self-hosted Chroma server
Optional: API key for external embedding providers (OpenAI, Cohere, Jina, VoyageAI, Roboflow)

Install

The server runs via uvx, no global install required:

uvx chroma-mcp

Claude Desktop config

Ephemeral (in-memory, dev only):

{
  "mcpServers": {
    "chroma": {
      "command": "uvx",
      "args": ["chroma-mcp"]
    }
  }
}

Persistent (local file storage):

{
  "mcpServers": {
    "chroma": {
      "command": "uvx",
      "args": [
        "chroma-mcp",
        "--client-type", "persistent",
        "--data-dir", "/path/to/data"
      ]
    }
  }
}

Chroma Cloud:

{
  "mcpServers": {
    "chroma": {
      "command": "uvx",
      "args": [
        "chroma-mcp",
        "--client-type", "cloud",
        "--tenant", "YOUR_TENANT_ID",
        "--database", "YOUR_DB_NAME",
        "--api-key", "YOUR_API_KEY"
      ]
    }
  }
}

HTTP (self-hosted Chroma):

{
  "mcpServers": {
    "chroma": {
      "command": "uvx",
      "args": [
        "chroma-mcp",
        "--client-type", "http",
        "--host", "your-host",
        "--port", "8000",
        "--ssl", "true"
      ]
    }
  }
}

Environment variables

CHROMA_CLIENT_TYPE: ephemeral | persistent | http | cloud
CHROMA_DATA_DIR: path for persistent storage
CHROMA_TENANT, CHROMA_DATABASE, CHROMA_API_KEY: Cloud credentials
CHROMA_HOST, CHROMA_PORT, CHROMA_SSL, CHROMA_CUSTOM_AUTH_CREDENTIALS: HTTP client settings
CHROMA_DOTENV_PATH: custom .env path (default .chroma_env)
CHROMA_<PROVIDER>_API_KEY: e.g. CHROMA_OPENAI_API_KEY, CHROMA_COHERE_API_KEY for embedding providers

CLI args take precedence over environment variables.

Use Cases

Give an AI agent persistent memory by storing conversation summaries in a Chroma collection and retrieving relevant past context on each turn.
Build a RAG knowledge base over internal docs: ingest with chroma_add_documents, query with chroma_query_documents for grounded answers.
Maintain a vector index of product descriptions or support tickets for semantic similarity search inside agent workflows.
Run a local persistent collection during development, then promote the same agent to Chroma Cloud for production without code changes.
Use metadata filters (where clauses) to scope searches by tenant, user, or document type in multi-user RAG systems.

Example Prompts

"Create a Chroma collection called support_docs and add these 20 FAQ entries with their categories as metadata."
"Search my engineering_notes collection for documents about retry logic and return the top 5 matches."
"How many documents are in the customer_feedback collection, and show me a sample of 3."
"Delete all documents in archive where metadata year equals 2022."
"List all my collections and tell me which one has the most documents."

Pros

Official server maintained by the Chroma team (chroma-core org).
Covers the full collection and document lifecycle with 12 well-scoped tools.
Flexible deployment: ephemeral, persistent local, self-hosted HTTP, or Chroma Cloud.
Pluggable embedding functions including OpenAI, Cohere, Jina, VoyageAI, and Roboflow.

Limitations

Requires Python and uvx on the host, which may be friction for non-Python environments.
Cloud and external embedding providers require separately managed API keys.
Ephemeral mode loses all data on restart, so production use requires Cloud or HTTP/persistent setup.

Alternatives

Pinecone MCP for managed vector search via Pinecone.
Qdrant MCP Server for Qdrant-backed vector storage and retrieval.
Weaviate MCP server for Weaviate-based hybrid search and RAG.