Chroma

You can run a Chroma-based MCP Server that provides vector storage, semantic search, and document management with persistent local storage. This server uses Chroma for embeddings and offers CRUD operations plus semantic retrieval to help you organize and search documents by meaning.

How to use

To operate the server, start it with the standard runtime command and then interact with it through MCP tools to create, read, update, delete, and search documents.

How to install

Prerequisites: you need Python 3.8 or later.

Create a local Python environment and install required tooling using the provided commands.

uv venv
uv sync --dev --all-extras

Additional configuration and usage notes

The server stores documents and metadata in a local data directory for persistent storage across restarts.

Start the server when ready, then use MCP client calls to manage documents and perform semantic searches.

Configuration for Claude Desktop shows how to point the client at the local server when running in a Windows environment.

{
  "mcpServers": {
    "chroma": {
      "command": "uv",
      "args": [
        "--directory",
        "C:/MCP/server/community/chroma",
        "run",
        "chroma"
      ]
    }
  }
}

Usage examples

Create a document and then search for semantically similar documents using metadata filters.

# Start the server in one terminal:
uv run chroma

# Create a document
create_document({
  "document_id": "ml_paper1",
  "content": "Convolutional neural networks improve image recognition accuracy.",
  "metadata": {
    "year": 2020,
    "field": "computer vision",
    "complexity": "advanced"
  }
})

# Search similar documents
search_similar({
  "query": "machine learning models",
  "num_results": 2,
  "metadata_filter": {
    "year": 2020,
    "field": "computer vision"
  }
})

Available tools

create_document

Create a new document with a unique document_id, content, and optional metadata.

read_document

Retrieve a document by its document_id, returning content and metadata.

update_document

Update the content and optional metadata of an existing document by its document_id.

delete_document

Delete a document by its document_id, removing it from storage.

list_documents

List all documents with their content and metadata, with optional limit and offset for pagination.

search_similar

Find semantically similar documents to a given query, with optional num_results and filters for metadata and content.