Vectra MCP Server

What is Vectra MCP Server?

A Model Context Protocol (MCP) server for managing and querying a Vectra knowledge base. It exposes tools to create collections, embed texts and files, and perform hybrid search (vector + keyword) with graph enhancement. Built in TypeScript for use with MCP-compatible clients.

How to use Vectra MCP Server?

Install dependencies with npm install, build with npm run build, and run the server (listens on stdio) with node build/index.js. For development with auto-rebuild, use npm run watch.

Key features of Vectra MCP Server

• Create and list Vectra collections.
• Embed text items or local files (batch) into Vectra.
• Query collections using hybrid search (vector + keyword) with optional graph enhancement.
• Add already embedded files to collections and list files in a collection.
• Delete files and their associated embeddings.
• Fetch nodes directly from the underlying ArangoDB database by key.

Use cases of Vectra MCP Server

• Build and maintain a searchable knowledge base from documents.
• Automate embedding of local files into Vectra collections.
• Enable LLM-powered retrieval-augmented generation (RAG) workflows via MCP clients.

FAQ from Vectra MCP Server

What tools does the server expose?

It provides tools for collection management, embedding texts/files, querying collections, file management, and direct ArangoDB node retrieval. See the README’s tools list for full input details.

How does querying work?

The query_collection tool always uses hybrid search (vector + keyword) with graph search enabled by default. You can optionally limit results, set a max distance, and filter by metadata or graph relationships.

What are the runtime requirements?

A Node.js environment (npm installed). The server listens on stdio; a Vectra backend API must be running separately.

What transport does the server use?

The server uses stdio transport (standard input/output), as is typical for MCP servers.

Where are embeddings and data stored?

Data lives in the Vectra backend and its underlying ArangoDB database. The server does not store data locally beyond configuration.