Quickstart

Install

Install from PyPI. The package name is mfs-cli; the command is mfs:

uv tool install mfs-cli
mfs --help

For one-off use without installing a persistent tool:

uvx --from mfs-cli mfs --help

For local embeddings or optional providers, install the matching extra:

uv tool install "mfs-cli[onnx]"
mfs config set embedding.provider onnx

For development from source:

git clone https://github.com/zilliztech/mfs.git
cd mfs
uv sync
uv run mfs --help

The default install uses OpenAI embeddings and Milvus Lite.

export OPENAI_API_KEY="sk-..."

For local embeddings while developing from source:

uv sync --extra onnx
mfs config set embedding.provider onnx

Build your first index

mfs add .

By default, mfs add scans files, writes work into ~/.mfs/queue.json, starts a detached worker, and returns quickly.

To wait for embedding in the foreground:

mfs add . --sync

Check progress:

mfs status

Search

Search needs an explicit path scope or --all.

mfs search "where do we configure database retries" .
mfs search "oauth callback flow" ./docs --top-k 5
mfs search "ERR_TOKEN" ./src --mode keyword
mfs search "session storage" --all

Browse

mfs tree --peek -L 2 .
mfs ls --skim ./docs
mfs cat --skim ./docs/auth.md
mfs cat -n 40:90 ./docs/auth.md

Exact search

mfs grep "ERR_TOKEN_EXPIRED" .
mfs grep -C 5 "OAuth" ./docs
mfs grep -i "refresh token" --all

Indexed files use Milvus-backed keyword filtering. Non-indexed text-like files can still be searched through the system grep fallback.