# vectorize-io/vectorize-mcp-server [Health: Active]

**Category:** 🔎 Search & Data Extraction  
**Repository:** https://github.com/vectorize-io/vectorize-mcp-server/  
**GitHub Stars:** 115  
**npm Downloads (last month):** 117  
**Views:** 5  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/vectorize-io-vectorize-mcp-server

## Description
Vectorize MCP server for advanced retrieval, Private Deep Research, Anything-to-Markdown file extraction and text chunking.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "vectorize-mcp-server": {
    "command": "npx",
    "args": ["-y","@vectorize-io/vectorize-mcp-server@latest"],
    "env": {
      "VECTORIZE_ORG_ID": "",
      "VECTORIZE_TOKEN": "",
      "VECTORIZE_PIPELINE_ID": ""
    }
  }
}
```

**Requires environment variables:** `VECTORIZE_ORG_ID`, `VECTORIZE_TOKEN`, `VECTORIZE_PIPELINE_ID` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What vectorize-io/vectorize-mcp-server MCP server does

The vectorize-io/vectorize-mcp-server MCP server exposes selected Vectorize capabilities to MCP-compatible AI clients. It is intended for workflows that need retrieval from a Vectorize pipeline, conversion of a document into Markdown, or private deep research based on pipeline data.

The server does not provide a local search index or document store of its own. Its operations depend on the Vectorize organization, access token, and pipeline configured at startup. That makes it a fit for agents that already have a Vectorize pipeline available and need to invoke it through MCP.

## How it works

The process is launched with the npm package through `npx`. At startup, it reads three environment variables: the Vectorize organization identifier, an authentication token, and the pipeline identifier. An MCP client then exposes the server's operations to an agent.

Retrieval accepts a natural-language question and a requested result count. The server passes that request to Vectorize's retrieval API and returns matching documents. Extraction accepts a Base64-encoded file and its MIME type, then uses Vectorize to produce Markdown-formatted text and chunks. Deep research accepts a query and a setting for web search, producing research from the configured pipeline.

## Setup and configuration

Install and run the vectorize-io/vectorize-mcp-server MCP server with:

```bash
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
```

The same command and variables can be placed in an MCP client configuration. The repository documents configuration for VS Code, Claude Desktop, Windsurf, Cursor, and Cline. VS Code configuration can be stored in user settings or in a workspace-level `.vscode/mcp.json` file. The token is represented as a password input in the supplied VS Code setup examples.

## Tools and capabilities

The vectorize-io/vectorize-mcp-server MCP server provides three documented operations:

- `retrieve` performs vector retrieval using a question and `k`, the requested number of results. For example, an agent can ask for the financial health of a company and request five results.
- `extract` accepts `base64document` and `contentType`. It extracts document text and returns it in Markdown chunks, supporting an “any file to Markdown” workflow as described by the project.
- `deep-research` accepts `query` and `webSearch`. It generates private deep research from the configured pipeline, with the web-search option allowing the request to include web search.

These operations cover separate stages of an agent workflow: finding relevant pipeline content, preparing file content for model use, and generating a research result.

## Limitations and notes

It also does not specify retrieval ranking behavior, maximum input sizes, response schemas, pipeline creation steps, or the exact scope of web search.

A valid Vectorize organization ID, token, and pipeline ID are required for the documented startup configuration. The server is configured to use one pipeline ID through the environment, so the material does not establish how an agent selects among multiple pipelines during a session. Client compatibility is documented through configuration examples rather than through a detailed compatibility matrix.

_Full upstream README: https://allmcps.com/mcp/vectorize-io-vectorize-mcp-server/readme_