Vectorize MCP Server for Vector Retrieval and Extraction

Overview

The Vectorize MCP Server is a Model Context Protocol (MCP) server implementation that connects MCP-enabled developer tools and LLM clients to Vectorize pipelines. It exposes endpoints for high-performance vector retrieval, document extraction and chunking, and pipeline-driven deep research — letting LLMs request contextual documents and structured content on demand.

This server is useful when you want to combine large language models with a scalable retrieval layer: index and search embeddings via Vectorize, extract text from arbitrary files, and run research-style queries (including optional web search) — all through a standardized MCP interface. It is designed to be simple to run locally or within developer environments like VS Code, Claude/Windsurf/Cursor, or other MCP-compatible clients.

Features

• Vector retrieval: k-nearest document search using Vectorize pipelines.
• Text extraction: convert PDFs, DOCX, and other documents into structured Markdown chunks.
• Deep research: create private multi-source research runs using Vectorize pipelines (supports optional web search).
• MCP-compatible: runs as an MCP server so LLM clients can call it as a tool.
• Lightweight deployment: start with a single npx command or embed in VS Code / workspace settings.
• Developer-friendly: local dev script, simple env-based configuration, and reproducible settings for teams.

Installation / Configuration

Prerequisites: node (npm) and a Vectorize account (Organization ID, Token, Pipeline ID).

Environment variables:

Quick start (npx)

export VECTORIZE_ORG_ID=your_org_id
export VECTORIZE_TOKEN=your_token
export VECTORIZE_PIPELINE_ID=your_pipeline_id

npx -y @vectorize-io/vectorize-mcp-server@latest

VS Code (user settings JSON)

{
  "mcp": {
    "inputs": [
      { "type": "promptString", "id": "org_id", "description": "Vectorize Organization ID" },
      { "type": "promptString", "id": "token", "description": "Vectorize Token", "password": true },
      { "type": "promptString", "id": "pipeline_id", "description": "Vectorize Pipeline ID" }
    ],
    "servers": {
      "vectorize": {
        "command": "npx",
        "args": ["-y", "@vectorize-io/vectorize-mcp-server@latest"],
        "env": {
          "VECTORIZE_ORG_ID": "${input:org_id}",
          "VECTORIZE_TOKEN": "${input:token}",
          "VECTORIZE_PIPELINE_ID": "${input:pipeline_id}"
        }
      }
    }
  }
}

Workspace (shareable .vscode/mcp.json)

{
  "inputs": [ /* same inputs as above */ ],
  "servers": {
    "vectorize": {
      "command": "npx",
      "args": ["-y", "@vectorize-io/vectorize-mcp-server@latest"],
      "env": {
        "VECTORIZE_ORG_ID": "${input:org_id}",
        "VECTORIZE_TOKEN": "${input:token}",
        "VECTORIZE_PIPELINE_ID": "${input:pipeline_id}"
      }
    }
  }
}

MCP client configuration (Claude/Windsurf/Cursor/Cline)

{
  "mcpServers": {
    "vectorize": {
      "command": "npx",
      "args": ["-y", "@vectorize-io/vectorize-mcp-server@latest"],
      "env": {
        "VECTORIZE_ORG_ID": "your-org-id",
        "VECTORIZE_TOKEN": "your-token",
        "VECTORIZE_PIPELINE_ID": "your-pipeline-id"
      }
    }
  }
}

Development

git clone https://github.com/vectorize-io/vectorize-mcp-server.git
cd vectorize-mcp-server
npm install
npm run dev

Release: update package.json version, commit, tag, and push tags.

Available Tools / Resources

The server exposes several MCP tools (sendable as tool calls from an LLM):

• retrieve — vector search; returns top-k matching documents. Example arguments:

  { "question": "Financial health of the company", "k": 5 }
  

• extract — extract and chunk text from a base64-encoded document into Markdown. Example arguments:

  { "base64document": "<base64>", "contentType": "application/pdf" }
  

• deep-research — run a private research pipeline that can include web search. Example arguments:

  { "query": "Generate a financial status report about the company", "webSearch": true }
  

Reference: the server uses Vectorize pipeline APIs (retrieval, extraction, and deep research). See Vectorize documentation for pipeline specifics and API contract.

Use Cases

• Retrieval-Augmented Generation (RAG): Connect an LLM to the retrieve tool to gather relevant document passages before answering user queries, improving factuality and context-awareness.
• Document ingestion pipeline: Upload PDFs/DOCs to Vectorize, use extract to produce Markdown chunks, and index them for fast retrieval in your application.
• Research assistant: Use deep-research to orchestrate multi-source research runs (internal docs + optional web search), producing summarized findings for analysts or product teams.
• Enterprise search: Power company-wide search by exposing a Vectorize-backed MCP server to internal LLMs and apps, enabling semantic search over knowledge bases, contracts, and reports.

Tips & Best Practices

• Keep tokens and IDs out of source control; prefer environment variables or secret management.
• Tune k and retrieval scoring in your Vectorize pipeline to match your domain’s document granularity.
• Use extraction chunk sizes that match your LLM context window for best end-to-end performance.
• Monitor Vectorize pipeline usage and pipeline logs to understand latency and throughput for production workloads.

Repository and source code: https://github.com/vectorize-io/vectorize-mcp-server/