What it does
The server ranks and filters retrieved context on the user's own machine, so that only the passages likely to answer a query are sent to the model. It runs on the AsterMind Community Edition ELM engine and provides ten tools, including rerank_documents, filter_context and compress_context, along with token counting and savings estimation.
Developers building RAG pipelines would reach for it when retrieval returns many chunks per prompt and they want to cut input tokens per call. In the publisher's benchmark across six scenarios, keeping the top 3 of 10 candidates reduced average context tokens by 66.9%, and a passage that answered the query survived filtering in every scenario. Strict recall of all relevant passages was 55.6%. Keeping fewer passages saves more but risks dropping a needed one, so the publisher suggests top 1 to 3 for single-hop questions and top 5 to 6 for multi-hop questions.
Running it requires Node 18 or later and an MCP client such as Claude Desktop, Cursor, VS Code or Cline, added through that client's configuration. According to the publisher, it makes no network calls, uses no API keys and keeps data on the machine.
Limitations and requirements
Requires Node 18 or later and an MCP client. The reranker is lexical (TF-IDF), so queries that use different wording from the answer may not rank it highly. Classification is weak and flagged as low confidence; embeddings are character-level and suited mainly to near-duplicate detection. Token savings hold only when a relevant passage is kept.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 13 Nov 2026 · its page on the source
AsterMind MCP
On-device reranking that cuts RAG context tokens ~67% while keeping a relevant passage.
v0.1.0As shown on registry.modelcontextprotocol.io when we last read it