Vectara MCP Server

What is Vectara MCP Server?

Vectara MCP Server provides any agentic application with access to fast, reliable RAG (Retrieval-Augmented Generation) with reduced hallucination, powered by Vectara’s Trusted RAG platform, through the Model Context Protocol (MCP). It is compatible with Claude Desktop and any MCP client.

How to use Vectara MCP Server?

Install via pip install vectara-mcp and run with python -m vectara_mcp. Configure your Vectara API key via the VECTARA_API_KEY environment variable or the setup_vectara_api_key tool. For Claude Desktop, use the STDIO transport in your MCP configuration.

Key features of Vectara MCP Server

• RAG query with generated answers via ask_vectara
• Semantic search without generation via search_vectara
• Hallucination detection and correction via correct_hallucinations
• Factual consistency evaluation via eval_factual_consistency
• Secure transport modes: HTTP, SSE, and STDIO
• API key management with one-time setup

Use cases of Vectara MCP Server

• Running RAG queries that combine search results with a generated answer
• Performing semantic search across Vectara corpora without response generation
• Detecting and correcting hallucinations in AI‑generated text
• Evaluating the factual consistency of generated text against source documents

FAQ from Vectara MCP Server

What does Vectara MCP Server do that alternatives don’t?

It provides Vectara’s dedicated RAG, hallucination correction, and factual consistency evaluation tools through the MCP standard, enabling any MCP‑compatible agent to access these capabilities securely.

What are the dependencies and runtime requirements?

Python 3.x, the vectara-mcp package from PyPI, and a valid Vectara API key. No other external services are required.

Where does data live?

All queries are processed by Vectara’s cloud platform. The server itself acts as a proxy; no data is stored locally by the MCP server.

What transport modes are supported?

HTTP (default, with authentication and rate limiting), SSE (streaming with bearer token support), and STDIO (for local development and Claude Desktop).

How is authentication handled?

By default, HTTP and SSE transports require a bearer token set via the VECTARA_API_KEY environment variable or the setup_vectara_api_key tool. STDIO transport uses the environment variable. Authentication can be disabled for development with --no-auth.