Which server supports more search modes?

Chroma's exposes semantic, metadata, and full-text search across collections, plus full collection management. LanceDB's server focuses on hybrid retrieval through three tools (catalog_search, chunks_search, all_chunks_search) over a local index, which is narrower but purpose-built for agentic RAG.

Can both run locally?

Yes. LanceDB's server points at a local index directory by design. Chroma's server offers persistent on-disk and ephemeral in-memory modes (as well as HTTP and Chroma Cloud), so either can run entirely on your machine without a hosted service.

LanceDB vs Chroma

LanceDB MCP and Chroma MCP both put a vector store behind an agent for retrieval, but they differ in scope and provenance. The LanceDB server (a community project, vendor Adiom) is a focused agentic-RAG tool over a local LanceDB index: it exposes catalog_search, chunks_search, and all_chunks_search so an agent can do hybrid search across a document catalog and its chunks — a ready-made retrieval layer for a local index directory. Chroma's server is official and broader: it manages collections (create, modify, delete, peek, count, get info) and runs semantic, metadata, and full-text search, with a backend mode you choose per run — ephemeral in-memory, persistent on disk, an HTTP server, or Chroma Cloud. So LanceDB's server is a lean, opinionated RAG-over-a-local-index helper, while Chroma's is a flexible collection-management and multi-mode search toolkit. Here is a balanced look at how they differ.

How they compare

Dimension	LanceDB	Chroma
Primary job	Agentic RAG over a local LanceDB index: hybrid search across a document catalog and its chunks.	Collection management plus semantic, metadata, and full-text search over a Chroma vector database.
Backend modes	Points at a local LanceDB index directory you supply — built for on-machine retrieval.	Pick a client type per run — ephemeral (in-memory), persistent (on disk), http, or Chroma Cloud — from scratchpad to managed.
Tool surface	Three retrieval tools (catalog_search, chunks_search, all_chunks_search) — narrow and RAG-oriented.	Many tools: collection lifecycle (create, modify, delete, peek, count, info) plus multiple search modes.
Official status	Community server (vendor Adiom), not published by LanceDB — weigh that provenance for support.	Official Chroma server, published by Chroma.
Best-fit task	Wiring an agent to do hybrid RAG over an existing local LanceDB index with minimal setup.	Managing multiple collections and combining semantic, metadata, and full-text search, with an easy zero-config local start.

Verdict

Both serve retrieval, so choose by scope and provenance. Pick the LanceDB server when you want a lean, opinionated agentic-RAG layer over a local LanceDB index — hybrid catalog and chunk search with little to configure — accepting that it is a community server. Pick the Chroma server when you want official, flexible collection management and multiple search modes (semantic, metadata, full-text), with a backend you choose per run from in-memory up to Chroma Cloud. In short: LanceDB for focused hybrid RAG over a local index; Chroma for official, multi-mode collection management and search with a frictionless local start.

FAQ

Which server supports more search modes?: Chroma's exposes semantic, metadata, and full-text search across collections, plus full collection management. LanceDB's server focuses on hybrid retrieval through three tools (catalog_search, chunks_search, all_chunks_search) over a local index, which is narrower but purpose-built for agentic RAG.
Can both run locally?: Yes. LanceDB's server points at a local index directory by design. Chroma's server offers persistent on-disk and ephemeral in-memory modes (as well as HTTP and Chroma Cloud), so either can run entirely on your machine without a hosted service.