MFS CLI
MFS gathers code, agent memory, documents and databases into one file-like tree searched by meaning and by text.
- 158 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- Indexes the contents of connected sources, including chat, mail and databases.
- Starts a local server and downloads an embedding model.
Install
- Works in
Manual install
uv tool install mfs-server && mfs-server runThe server, no clone needed. Wait for `Application startup complete`, then install the client: `cargo install mfs-cli`.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Indexes the contents of connected sources, including chat, mail and databases.
- Starts a local server and downloads an embedding model.
This is third-party code. Review the repository files before installing.
What it does
MFS (Multi-source File-like Search) consists of an `mfs-server` server, an `mfs` command-line client and two agent skills. Every source becomes a tree under a stable address: repositories, memory and skill files, PDF and Word documents, chat and mail, issues, databases and object stores. The `mfs-ingest` skill bundles connection commands: `mfs add` registers and indexes a source, `mfs connector` lists and removes connections. The `mfs-find` skill searches with `search` and `grep` and opens results with `ls`, `cat`, `tree`, `head` and `tail` down to exact bytes. Documents convert to text locally; it runs on Milvus.
Who it is for. Agent developers and engineers with scattered context who want one search across repositories, notes and documents.
Good fit when
- An agent looks through past sessions for a decision.
- A function must be found by meaning in a large repository.
- Several projects and documents are indexed into one tree.
- Context lives in Slack, Jira and a database and the agent needs a shared entry point.
Not a fit when
- Plain `grep` over one folder is enough.
- The machine has no room for a local model of about 600 MB.
- Sources with private data cannot be indexed on this machine.
Example request
Use the mfs-ingest skill to index the notes folder and the repository, then use mfs-find to locate where we settled the request retry question.Limitations
A running server is required: `uv tool install mfs-server && mfs-server run` or Docker Compose, plus the client via `cargo install mfs-cli`. The first run downloads a local embedding model of about 600 MB into `~/.mfs/`. With an embedding service key the server can be pointed at it instead. Connecting chat, mail and databases indexes their contents, so separate access. On macOS, clear quarantine from `mfs` if warned about an unknown developer. The last commit in the repository dates from July 2026.
How to disable. Remove the skills from the agent's directory, stop `mfs-server` and delete `~/.mfs/` with the index and model. Remove connections with `mfs connector`.
FAQ
What parts does MFS have?
The `mfs-server` server, the Rust `mfs` client and the `mfs-ingest` and `mfs-find` skills.
How do I install the skills for Claude Code only?
Run `npx skills add zilliztech/mfs -a claude-code -a codex -g`; for a single agent keep just the relevant `-a` flag.
Is an API key required?
Not in local mode: the model and document conversion run on the machine. Image descriptions need a vision-capable model.
How do I check it works?
Create a folder with a file, run `mfs add --upload <folder>` and `mfs search "query" <folder>`.
What gets indexed besides files?
Connectors bring in chat, mail, issues, databases and object stores. List connections with `mfs connector`.
Discussion
No questions or reviews yet.
Related
CLI toolsA local MCP server that turns a folder of documents into a knowledge base with hybrid search for Claude Code and Cursor
A skill that teaches an agent to search tables, trace lineage and read the glossary in your Marmot catalog via CLI, REST and MCP.
An MCP server gives agents search and answers over local documents using hybrid retrieval, reranking and section links.
A knowledge graph with an MCP server: the agent searches concepts by meaning and sees how well each is supported.
Elasticsearch MCP connects multiple search clusters, with an explicit target name and access scoped to the supplied credentials.
MCP server for the Flexible GraphRAG platform: ingest from 14 data sources, hybrid search and AI answers over a knowledge graph
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| MFS (this card) | no | 158 | ||
| memsearch | no | 2,868 | search over agent memory; MFS adds repositories, documents, mail and databases and works through skills with file commands | |
| Claude Context | yes | API key | 12,690 | indexes code for semantic search; MFS gathers code together with memory, documents and external sources |
README in short
MFS gathers agent context from many sources into one file-like space and hands it to the `mfs-ingest` and `mfs-find` skills. It runs locally on Milvus.