Skip to content

Get started with MCP＃

Connect kluster.ai's verification tools to your AI assistant through Model Context Protocol (MCP). This guide shows you how to enable Cloud MCP and integrate it with Claude desktop for real-time claim validation directly within your conversations.

Cloud MCP provides managed verification endpoints with no infrastructure to maintain - just enable your MCP endpoint and start verifying.

Prerequisites＃

Before getting started, ensure you have:

  • A kluster.ai account: Sign up on the kluster.ai platform if you don't have one.
  • A kluster.ai API key: After signing in, go to the API Keys section and create a new key. For detailed instructions, check out the Get an API key guide.
  • Claude desktop for testing the integration.

Enable MCP＃

To enable the MCP endpoint, go to the kluster.ai platform and take the following steps:

  1. Navigate to MCP and view your current MCP status.
  2. Click the Enable Verify MCP button to activate your endpoint.
  3. Copy your client configuration.

Endpoint enabled

Your MCP endpoint is now active. Copy your API key and save it securely.

The platform provides ready-to-use integration examples for VSCode, Cursor, Claude code, and Claude desktop.

MCP kluster.ai platform

Configure Claude desktop＃

Edit your Claude desktop configuration file:

~/Library/Application Support/Claude/claude_desktop_config.json

%APPDATA%/Claude/claude_desktop_config.json

Add the MCP server configuration:

{
    "mcpServers": {
        "kluster-verify-mcp": {
            "command": "npx",
            "args": [
                "mcp-remote",
                "https://api.kluster.ai/v1/mcp",
                "--header",
                "Authorization: Bearer YOUR_MCP_TOKEN"
            ]
        }
    }
}

Replace YOUR_MCP_TOKEN with your actual token or copy the snippet from the platform.

Restart Claude desktop. Once Claude desktop restarts, you'll see the verification tools available under kluster-verify-mcp.

List tools on Claude desktop

Available tools＃

Your MCP integration provides two verification tools:

  • verify: Validates prompt and response pairs against reliable sources.
  • verify_document: Verifies prompt and response pairs in relation to uploaded documents.

Verify＃

Ask Claude to verify something obviously wrong:

"The Eiffel Tower is located in Rome. Use the verify tool to check this."

Claude will automatically use the verify tool with:

  • prompt: "Is the Eiffel Tower located in Rome?"
  • response: "The Eiffel Tower is located in Rome."

And provides the following:

  • Verification result: Whether the response contains hallucinations.
  • Detailed explanation: Why it's wrong with supporting reasoning.
  • Source citations: Search results used for verification.

Verify MCP tool demo

Verify documents＃

Perfect for detecting hallucinations or false claims about documents. Upload any document to Claude, then ask:

"Does this document say that employees can work remotely full-time? The document says employees can work remotely without restrictions. Use the verify_document tool to check."

Claude will use the verify_document tool with:

  • prompt: "Does this document say that employees can work remotely full-time?"
  • response: "The document says employees can work remotely without restrictions."
  • documentContent: The content of the document as provided by the MCP client for verification.

This verifies the response against the actual document content.

Alternative setup options＃

  • Other clients: Want to use VS Code, Cursor, or Claude Code? Check the Client integrations guide for configuration examples.

  • Self-hosted: Prefer to run MCP locally? Set up the self-hosted MCP server for local development with full control.

  • API activation: Enable MCP using API calls with the MCP API usage guide.

Next steps＃