Summary
Your own indexed corpus answers, rather than the model's memory.
retrieve pulls the passages behind an answer out of the pipeline you already built, so what supports a claim is a document you own. extract runs the other direction — a file arrives base64-encoded and comes back as chunks — and deep-research takes a question broad enough that one retrieval would not settle it.
What it is
A client for a configured Vectorize pipeline: retrieval of the documents that answer a question, text extraction and chunking of a document you pass in, and a deep-research run over the same pipeline.
What you get
- Documents retrieved from the configured pipeline for a question, with k setting how many come back
- Text extracted and chunked from a base64-encoded document of a content type you declare
- A deep research run generated over the pipeline, optionally widened with web search
Requirements
A Vectorize organisation and pipeline — VECTORIZE_ORG_ID and VECTORIZE_PIPELINE_ID — with a token in VECTORIZE_TOKEN, plus npx on your PATH.
Setup effort
One command plus a key — npx -y @vectorize-io/vectorize-mcp-server@latest, then supply credentials