Skip to content
MCP server

mcp-vl-msa-rs

Local BM25 corpus recall and memory tools for MCP agents, written in Rust

Pricing
Open source
Platforms
Linux, macOS, Android
Version
0.4.1
Listing updated
today

Similar MCP servers

About mcp-vl-msa-rs

mcp-vl-msa-rs is an open-source MCP server for knowledge and memory. It is available on Linux, macOS, Android, CLI and self-hosted.

The README describes this server as long-term memory for AI agents, run over MCP stdio. Documents, notes and past conversations are indexed into named collections, and a query returns the top matching chunks along with the original text, so a model gets exact passages rather than only embedding matches.

Retrieval is built on tantivy BM25 and runs without a cloud account, embedding service or ML dependencies in the default build. An optional in-process dense rerank can be enabled with a feature flag and a local model bundle, and the per-call dense_alpha parameter controls how much weight it gets.

The project is aimed at developers who run coding agents such as Claude Code or Codex and want recall over a docs tree or their own chat history. The README positions it as the corpus half of a pair, with the companion mcp-memory-rs handling curated state.

Tools this server provides (11)

Taken from the project documentation; the running server may differ.

msa_index
Index a document; existing chunks for doc_id are replaced.
msa_search
Top-k chunks, score normalized 0.0–1.0.
msa_fetch_doc
Full original text of a document.
msa_delete
Remove a document and all its chunks.
msa_list_collections
Collections open in the registry.
msa_stats
Per-collection statistics.
msa_search_iterative
Memory Interleave search with server-side cursor.
msa_drop_session
Force-evict a Memory Interleave session before TTL.
msa_remember
Add an agent memory with enrich and dedup.
msa_forget
Remove an agent memory.
msa_sync_path
Mirror a directory into a collection.

What it does

BM25 corpus search
Indexes documents into collections and returns top-k chunks with normalized scores.
Original text injection
Fetches the full original text of a document so the model sees source passages, not only snippets.
Agent memory capsules
Adds and removes agent memories through remember and forget tools with content-hash dedup.
Optional hybrid rerank
Blends BM25 with in-process dense cosine scores when built with the embeddings feature.
Iterative multi-hop retrieval
Supports a server-side cursor so a client can run repeated search rounds across a session.
Directory sync
Mirrors a filesystem directory into a collection using blake3 delta sync.

Good for

  • Giving a coding agent recall over a project's docs tree
  • Letting an assistant search its own past conversation history
  • Querying a local knowledge base without sending data to a cloud service
  • Keeping agent memory across sessions and model changes

Pricing

Open source

Apache-2.0 licensed, free to self-host.

Prices change. Check dev.mmmbuto.com before you buy.

mcp-vl-msa-rs FAQ

Does it need an embedding service or cloud account?

The README says the default build uses pure BM25 with no network dependencies and no embedding service required. Dense rerank is optional and runs in-process from a local model bundle.

Which MCP clients does it work with?

The README names Claude Code and Codex and says any MCP client that speaks stdio can use it.

Which platforms have prebuilt binaries?

Prebuilt targets are listed for Linux and Android, including aarch64 Termux builds. The README says no prebuilt macOS binary is shipped and recommends building from source with cargo.

Is mcp-vl-msa-rs free?

Yes. mcp-vl-msa-rs is open source, so you can use and self-host it for free.

What are the best alternatives to mcp-vl-msa-rs?

The closest alternatives to mcp-vl-msa-rs are mcp-memory-rs, MCP Memory, Holographic Memory, Agent Memory MCP and Cross-Agent Memory.

Where this comes from

  • Description: written from the project's README, the project's website and its MCP Registry entry.

How AIPoweredBot checks servers