Summary
Company answers come from the company's own documents.
A single tool is the whole design: a query goes in, the passages Ragie holds come back, and the assistant answers from those rather than from training data. recencyBias and rerank address the case that defeats plain retrieval — a superseded policy that still matches the words better than the current one.
What it is
One retrieval tool against a Ragie knowledge base, with a filter, a topK, reranking and a recency bias.
What you get
- Documents and passages retrieved from the knowledge base for a query
- Results narrowed by filter and capped by topK
- Reranking applied to what comes back, and recencyBias to favour newer material
Requirements
A Ragie account with an API key in `RAGIE_API_KEY`, and a knowledge base with documents already in it. A partition id can be supplied when the knowledge base is split into several. npx on your PATH.
Setup effort
One command plus a key — RAGIE_API_KEY=your_api_key npx @ragieai/mcp-server, then supply credentials