vectorize

This MCP server integrates Vectorize to enable vector-based retrieval and text extraction as part of your MCP workflow. The server exposes tools for vector search (retrieve), document text extraction and chunking (extract), and advanced research capabilities (deep-research) that leverage your Vectorize pipeline. To use it, start the server with the configured environment variables (ORG ID, API token, and pipeline ID) and then issue the predefined MCP tool requests in your client or workspace. The retrieve tool will search your Vectorize index and return relevant documents, the extract tool can convert documents into Markdown-formatted text chunks, and deep-research can generate focused, research-oriented responses using your pipeline and optional web search.

Prerequisites:

• Node.js (LTS) and npm installed on your machine
• Access to a Vectorize account with an Organization ID, API Token, and a Pipeline ID

Installation steps:

1. Install Node.js and npm from https://nodejs.org/ if not already installed.
2. Open your terminal and set your Vectorize credentials as environment variables (replace placeholders with your actual values):

export VECTORIZE_ORG_ID=YOUR_ORG_ID
export VECTORIZE_TOKEN=YOUR_TOKEN
export VECTORIZE_PIPELINE_ID=YOUR_PIPELINE_ID

3. Run the MCP server using npx:

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

This will start the Vectorize-based MCP server configured to read the environment variables above. You can also install and pin the package within your project using npm if you prefer:

npm install -D @vectorize-io/vectorize-mcp-server@latest

4. If you use VS Code, you can paste the provided MCP JSON configuration into your user settings or workspace settings as shown in the README to wire the server with your environment inputs.

Tips:

• Ensure VECTORIZE_ORG_ID, VECTORIZE_TOKEN, and VECTORIZE_PIPELINE_ID are kept secret and not checked into version control.
• The server exposes three tools: retrieve (vector search for documents), extract (text extraction and Markdown chunking), and deep-research (private deep research via your pipeline). Each tool requires a specific JSON payload structure as shown in the README.
• If you encounter authentication errors, recheck token scopes in Vectorize and verify that the pipeline ID belongs to the same organization.
• When deploying in different environments (local, CI, or VS Code), consider exporting the environment variables in the respective environment startup script.
• You can override the default commands or add additional environment variables if your Vectorize setup requires extra configuration.