MCP Server Overview
The Vectorize Model Context Protocol (MCP) Server enables AI assistants and agents to securely retrieve structured, metadata-rich information from your pipelines. MCP powers workflows like retrieval-augmented generation (RAG), document extraction, and multi-step reasoning using your private data.
This page explains the two ways to use MCP with Vectorize:
- Local MCP Server: A command-line server you run on your machine.
- Remote MCP Server: A hosted service managed by Vectorize.
What Is MCP?
The Model Context Protocol (MCP) is an open standard designed to let AI agents access external tools and data sources. It defines:
- How agents send requests (e.g., retrieval, extraction).
- How servers respond with relevant, structured data.
With Vectorize, MCP enables assistants like Claude to:
- Query your private pipelines.
- Perform document extraction.
- Generate research reports.
MCP Server Options
| Feature | Local MCP Server | Remote MCP Server |
|---|---|---|
| Hosting | Run locally using npx | Hosted by Vectorize |
| Connection Method | Local process | Server-Sent Events (SSE) |
| Tool Configuration | Static via environment variables | Dynamic via Vectorize UI |
| Custom Tools | Not supported | Supported: Define functions per agent |
| Best For | Development, testing, small-scale use | Production-scale, multi-agent setups |
Choosing Between Local and Remote MCP
-
Choose Local MCP Server if:
- You want a lightweight setup for testing.
- You don’t need complex tool configuration.
-
Choose Remote MCP Server if:
- You need to manage multiple agents.
- You want to define custom tools and parameters.
- You prefer not to run local infrastructure.
Where to Go Next
For more details about MCP-compatible clients and configuration options, refer to the API Reference.