Skip to main content
MCP server CX Codex

Chroma — Claude Code MCP server: vector search and memory

Chroma gives an agent a vector store: collections, semantic search, metadata filters. Useful when context needs to persist across sessions.

no license updated 2026-10-29

Open on GitHub

What Chroma does

Chroma hooks an open-source embedding database into your agent. The model creates collections from generated data and user input, then pulls from them with vector search, full text search, or metadata filters.

The server exposes tools like chroma_create_collection, chroma_add_documents, chroma_query_documents, chroma_get_documents, chroma_update_documents, and chroma_delete_documents, plus a few for collection management. You can run it in-memory for tests, on disk for persistence, over HTTP against a self-hosted instance, or in cloud mode that connects to api.trychroma.com.

Embedding functions that call external APIs (cohere, openai, jina, voyageai, roboflow) require an API key in an environment variable. The exact format depends on the provider. One catch: embedding function persistence arrived in Chroma 1.0.0, so collections made on version 0.6.3 or older do not support it. Launch the server with uvx chroma-mcp.

What it helps with

  • give an agent memory between sessions
  • find documents by meaning instead of exact word match
  • filter query results by metadata
  • store data locally with no external service
  • update and remove documents from collections

Example prompts for the agent

  • Create a collection called notes and add these entries with date metadata
  • Search the docs collection for pages similar to "setting up CI"
  • How many documents are in the emails collection right now?

How to install Chroma

Install

The MCP registry has no launch command for it: see the repository README

Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more

Questions about Chroma

How do I install Chroma in Claude Code?

Add this to claude_desktop_config.json: "chroma": {"command": "uvx", "args": ["chroma-mcp"]} for an in-memory client. For disk storage, append --client-type persistent --data-dir /path/to/dir to args.

What does Chroma do?

It gives an agent access to an embedding database: create collections, add documents with metadata, and search them by vectors or full text. Think of it as long-term memory or a shared knowledge base for a model.

Which embedding functions does it support?

default, cohere, openai, jina, voyageai, and roboflow. External APIs need an environment variable with the right key. The chosen function sticks to a collection starting with Chroma 1.0.0.

Similar MCP servers

All MCP servers for Agent setup: memory, context, orchestration →

Project sponsors

Companies that keep this analytics open