Vectara MCP server

Vectara MCP Server is an implementation of the Model Context Protocol (MCP) that enables AI systems to interact with Vectara's Trusted RAG platform. It provides agentic applications with access to fast, reliable RAG capabilities with reduced hallucination risk, all through the standardized MCP protocol.

Installation

You can install the package directly from PyPI:

pip install vectara-mcp

Server Configuration

Starting the Server

Secure HTTP Mode (Default)

# Start server with secure HTTP transport (DEFAULT)
python -m vectara_mcp
# Server running at http://127.0.0.1:8000 with authentication enabled

Local Development Mode (STDIO)

# For Claude Desktop or local development
python -m vectara_mcp --stdio
# ⚠️ Warning: STDIO transport is less secure. Use only for local development.

Configuration Options

# Custom host and port
python -m vectara_mcp --host 0.0.0.0 --port 8080

# SSE transport mode
python -m vectara_mcp --transport sse --path /sse

# Disable authentication (DANGEROUS - dev only)
python -m vectara_mcp --no-auth

Transport Modes

HTTP Transport (Default - Recommended)

• Built-in authentication via bearer tokens
• HTTPS ready
• Rate limiting: 100 requests/minute by default
• Configurable CORS protection
• Ideal for: Production deployments, cloud environments

SSE Transport

• Server-Sent Events for real-time updates
• Bearer token support
• Works with legacy MCP clients
• Ideal for: Real-time streaming applications

STDIO Transport

• ⚠️ No transport-layer security
• Low latency for local communication
• Ideal for: Local development, Claude Desktop
• Must be explicitly enabled with --stdio flag

Environment Variables

# Required
export VECTARA_API_KEY="your-api-key"

# Optional
export VECTARA_AUTHORIZED_TOKENS="token1,token2"  # Additional auth tokens
export VECTARA_ALLOWED_ORIGINS="http://localhost:*,https://app.example.com"
export VECTARA_TRANSPORT="http"  # Default transport mode
export VECTARA_AUTH_REQUIRED="true"  # Enforce authentication

Authentication

HTTP/SSE Transport

When using HTTP or SSE transport, authentication is required by default:

# Using curl with bearer token
curl -H "Authorization: Bearer $VECTARA_API_KEY" \
     -H "Content-Type: application/json" \
     -X POST http://localhost:8000/call/ask_vectara \
     -d '{"query": "What is Vectara?", "corpus_keys": ["my-corpus"]}'

# Using X-API-Key header (alternative)
curl -H "X-API-Key: $VECTARA_API_KEY" \
     http://localhost:8000/sse

Disabling Authentication (Development Only)

# ⚠️ NEVER use in production
python -m vectara_mcp --no-auth

Available Tools

API Key Management

setup_vectara_api_key

Configure and validate your Vectara API key for the session (one-time setup).

Arguments:

• api_key: str, Your Vectara API key - required.

Returns:

• Success confirmation with masked API key or validation error.

clear_vectara_api_key

Clear the stored API key from server memory.

Returns:

• Confirmation message.

Query Tools

ask_vectara

Run a RAG query using Vectara, returning search results with a generated response.

Arguments:

• query: str, The user query to run - required.
• corpus_keys: list[str], List of Vectara corpus keys to use for the search - required.
• n_sentences_before: int, Number of sentences before the answer to include in the context - optional, default is 2.
• n_sentences_after: int, Number of sentences after the answer to include in the context - optional, default is 2.
• lexical_interpolation: float, The amount of lexical interpolation to use - optional, default is 0.005.
• max_used_search_results: int, The maximum number of search results to use - optional, default is 10.
• generation_preset_name: str, The name of the generation preset to use - optional, default is "vectara-summary-table-md-query-ext-jan-2025-gpt-4o".
• response_language: str, The language of the response - optional, default is "eng".

Returns:

• The response from Vectara, including the generated answer and the search results.

search_vectara

Run a semantic search query using Vectara, without generation.

Arguments:

• query: str, The user query to run - required.
• corpus_keys: list[str], List of Vectara corpus keys to use for the search - required.
• n_sentences_before: int, Number of sentences before the answer to include in the context - optional, default is 2.
• n_sentences_after: int, Number of sentences after the answer to include in the context - optional, default is 2.
• lexical_interpolation: float, The amount of lexical interpolation to use - optional, default is 0.005.

Returns:

• The response from Vectara, including the matching search results.

Analysis Tools

correct_hallucinations

Identify and correct hallucinations in generated text using Vectara's VHC (Vectara Hallucination Correction) API.

Arguments:

• generated_text: str, The generated text to analyze for hallucinations - required.
• documents: list[str], List of source documents to compare against - required.
• query: str, The original user query that led to the generated text - optional.

Returns:

• JSON-formatted string containing corrected text and detailed correction information.

eval_factual_consistency

Evaluate the factual consistency of generated text against source documents using Vectara's dedicated factual consistency evaluation API.

Arguments:

• generated_text: str, The generated text to evaluate for factual consistency - required.
• documents: list[str], List of source documents to compare against - required.
• query: str, The original user query that led to the generated text - optional.

Returns:

• JSON-formatted string containing factual consistency evaluation results and scoring.

Note: API key must be configured first using setup_vectara_api_key tool or VECTARA_API_KEY environment variable.

Configuration with Claude Desktop

To use with Claude Desktop, update your configuration to use STDIO transport:

{
  "mcpServers": {
    "Vectara": {
      "command": "python",
      "args": ["-m", "vectara_mcp", "--stdio"],
      "env": {
        "VECTARA_API_KEY": "your-api-key"
      }
    }
  }
}

Or using uv:

{
  "mcpServers": {
    "Vectara": {
      "command": "uv",
      "args": ["tool", "run", "vectara-mcp", "--stdio"]
    }
  }
}

Note: Claude Desktop requires STDIO transport. While less secure than HTTP, it's acceptable for local desktop use.

Usage in Claude Desktop App

After completing the installation and configuring the Claude desktop app, completely close and re-open the app to see the Vectara-mcp server. You should see a hammer icon in the bottom left of the app indicating available MCP tools. Click on the hammer icon to see more detail on the Vectara tools.

Claude will have access to all six Vectara tools.

Secure Setup Workflow

First-time setup (one-time per session):

1. Configure your API key securely:

setup-vectara-api-key
API key: [your-vectara-api-key]

After setup, use any tools without exposing your API key:

Vectara Tool Examples

1. RAG Query with Generation:

ask-vectara
Query: Who is Amr Awadallah?
Corpus keys: ["your-corpus-key"]

2. Semantic Search Only:

search-vectara
Query: events in NYC?
Corpus keys: ["your-corpus-key"]

3. Hallucination Detection & Correction:

correct-hallucinations
Generated text: [text to check]
Documents: ["source1", "source2"]

4. Factual Consistency Evaluation:

eval-factual-consistency
Generated text: [text to evaluate]
Documents: ["reference1", "reference2"]

Security Best Practices

1. Always use HTTP transport for production - Never expose STDIO transport to the network
2. Keep authentication enabled - Only disable with --no-auth for local testing
3. Use HTTPS in production - Deploy behind a reverse proxy with TLS termination
4. Configure CORS properly - Set VECTARA_ALLOWED_ORIGINS to restrict access
5. Rotate API keys regularly - Update VECTARA_API_KEY and VECTARA_AUTHORIZED_TOKENS
6. Monitor rate limits - Default 100 req/min, adjust based on your needs

How to install this MCP server

For Claude Code

To add this MCP server to Claude Code, run this command in your terminal:

claude mcp add-json "Vectara" '{"command":"uv","args":["tool","run","vectara-mcp"]}'

See the official Claude Code MCP documentation for more details.

For Cursor

There are two ways to add an MCP server to Cursor. The most common way is to add the server globally in the ~/.cursor/mcp.json file so that it is available in all of your projects.

If you only need the server in a single project, you can add it to the project instead by creating or adding it to the .cursor/mcp.json file.

Adding an MCP server to Cursor globally

To add a global MCP server go to Cursor Settings > Tools & Integrations and click "New MCP Server".

When you click that button the ~/.cursor/mcp.json file will be opened and you can add your server like this:

{
    "mcpServers": {
        "Vectara": {
            "command": "uv",
            "args": [
                "tool",
                "run",
                "vectara-mcp"
            ]
        }
    }
}

Adding an MCP server to a project

To add an MCP server to a project you can create a new .cursor/mcp.json file or add it to the existing one. This will look exactly the same as the global MCP server example above.

How to use the MCP server

Once the server is installed, you might need to head back to Settings > MCP and click the refresh button.

The Cursor agent will then be able to see the available tools the added MCP server has available and will call them when it needs to.

You can also explicitly ask the agent to use the tool by mentioning the tool name and describing what the function does.

For Claude Desktop

To add this MCP server to Claude Desktop:

1. Find your configuration file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json
• Linux: ~/.config/Claude/claude_desktop_config.json

2. Add this to your configuration file:

{
    "mcpServers": {
        "Vectara": {
            "command": "uv",
            "args": [
                "tool",
                "run",
                "vectara-mcp"
            ]
        }
    }
}

3. Restart Claude Desktop for the changes to take effect