create_documentLABSCO SUMMARY
A local memory you can inspect.
It is CRUD and similarity search over a store on your own disk: documents keep the metadata you gave them, results come back with distances rather than as an opaque ranking, and the whole thing is a directory you can copy or delete.
What it is
A document store backed by Chroma's vector database. Documents go in with an id, content and optional metadata; retrieval is semantic similarity with metadata and content filters on top; and the data persists in a local directory between restarts.
What you get
- Documents created with an id, content and key-value metadata
- A document read, updated or deleted by id
- The whole set listed, with limit and offset
- Semantic search returning a ranked list with distance scores
- Search narrowed by metadata fields and by document content
- Storage that persists in a local directory across restarts
Requirements
Python 3.8 or newer and a recent Chroma, installed with uv. Data is written to a directory inside the project, so the store is local and the backups are yours. No account and no key.
Setup effort
Build from source — git clone https://github.com/privetin/chroma
USE CASES
01
Giving an assistant a body of notes it can search by meaning
Add the documents with metadata, then search by what you meant rather than by the words you used.
02
Narrowing a search to one slice of the collection
Filter on the metadata you set when writing - project, source, date - so similarity is applied inside the right subset.
03
Keeping a stored document current
Read it by id and update it in place, and the search moves with it.
04
Auditing what the store actually holds
List the documents with a limit and offset, so a growing collection stays reviewable.
05
Removing something that should not be retrievable
Delete it by id - the store is on your disk, so the removal is something you can check.
6 tools
Listed in project docs. Your client may show a different list.
read_documentupdate_documentdelete_documentlist_documentssearch_similarBEFORE YOU INSTALL
What to know first
Tools you get
6 tools named in the project's documentation.
Requires
Python 3.8+
Where it runs
The documented client runs the server on your machine.
INSTALLATION
Connect your client
Server setup
TERMINAL
git clone https://github.com/privetin/chroma
cd chroma
uv venv
uv sync --dev --all-extrasMCP Clients
Claude Desktop
CLAUDE DESKTOP · JSON
{
"mcpServers": {
"chroma": {
"command": "uv",
"args": [
"--directory",
"C:/MCP/server/community/chroma",
"run",
"chroma"
]
}
}
}C:/MCP/server/community/chroma is the README's folder; yours is the full path of the chroma folder your git clone created. The config file is ~/Library/Application Support/Claude/claude_desktop_config.json on macOS, %APPDATA%\Claude\claude_desktop_config.json on Windows.
