Skip to content
Download Longwave

Chroma

Chroma is an MCP server for storing and searching embeddings in Chroma. Its connected implementation exposes 0 tools when the Chroma runtime is available.

Add to Longwave Source

Install elsewhere

Install Chroma

Choose a harness. These instructions start the same local MCP server shown on this page.

Claude Code

Run this in your terminal. It adds a user-scoped server; use --scope project instead if it should apply only to the current project.

terminal 
claude mcp add --scope user --transport stdio chroma -- uvx chroma-mcp

What it does

Chroma MCP gives an agent a vector database connection for retrieval over documents. It is the missing data layer when a task needs semantic search rather than a keyword grep.

  • Inspect Chroma collections.
  • Store documents and embeddings.
  • Search a collection for relevant content.
  • Use metadata and query arguments returned by the server.
  • Keep the Chroma connection in the server runtime.

Questions people ask

Can Chroma MCP search a vector database?

Yes. The server is for searching Chroma collections with embeddings and metadata. The live functions define the collection and query fields supported by this version.

Can Chroma MCP store embeddings?

The entry covers document and embedding operations in Chroma. Use the live function descriptions to confirm the write operation and metadata fields before loading a collection.

How do I set up Chroma MCP?

Install `uv`, add the catalog-backed server to your MCP client and configure the Chroma environment it should use. Start it on the machine that can reach the collection.

Can Chroma MCP run locally?

Yes. The server can run locally when the Chroma runtime and collection are available to that process. Keep the local data path or service connection in the server environment.

Can Chroma MCP work with Python?

The catalog uses `uv` for the server runtime. Let `uv` resolve the Python environment rather than relying on a different global interpreter.

Does Chroma MCP need a separate database?

It needs access to the Chroma environment configured for the server. That can be a local Chroma runtime or a reachable hosted setup, depending on how you deploy it.

Put your documents behind semantic search.

Give the agent retrieval over the collection instead of another giant context dump.

Download Longwave