Vectara

You can deploy and use the Vectara MCP Server to connect AI agents with data sources via the Model Context Protocol. It provides secure, two-way access to fast, reliable retrieval and generation capabilities, with authentication and configurable transports for local development or production deployments.

How to use

Start by running the MCP server in HTTP mode for production or SSE-compatible streaming, or use STDIO for local development with Claude Desktop or similar clients. The server handles authentication by default and exposes a request surface you can invoke from your MCP client to perform generation, semantic search, and quality checks across your Vectara corpora.

From your MCP client, connect to the server at the base URL and invoke the available endpoints to perform actions such as asking questions with generation, performing semantic searches, and using tools to manage API keys and run RAG workflows. Ensure your API key is configured in the environment or via the provided setup tooling before making requests.

How to install

Prerequisites: Python must be installed on your system. You can verify Python is available by running python --version.

Install the MCP package from PyPI.

pip install vectara-mcp

Configuration and usage notes

Secure by default: the server runs over HTTP/S with authentication enabled unless you opt into STDIO for local development.

Transport modes you can use include HTTP (default), SSE for streaming, and STDIO for local development with Claude Desktop. Each mode has its own security and usage considerations.

Environment variables you may configure include the API key and optional authentication controls to tailor access. The recommended setup uses the VECTARA_API_KEY in your environment.

For local development in Claude Desktop, configure STDIO transport to run the server directly from your client environment.

Available tools

setup_vectara_api_key

Configure and validate your Vectara API key for the session (one-time setup). Returns success confirmation or a validation error.

clear_vectara_api_key

Clear the stored API key from server memory. Returns a confirmation message.

ask_vectara

Run a RAG query using Vectara, returning search results with a generated response. Supports corpus keys, context framing, and generation presets.

search_vectara

Run a semantic search query using Vectara, returning matching search results without generation.

correct_hallucinations

Identify and correct hallucinations in generated text using Vectara's VHC API. Returns corrected text and correction details.

eval_factual_consistency

Evaluate the factual consistency of generated text against source documents using a dedicated API. Returns evaluation results and scoring.