Skip to main content

Deploy MCP

An MCP (Model Context Protocol) server gives AI assistants direct access to your document collection. Once deployed on your data cluster, AI tools like Claude Desktop can search, read, and analyze your documents through structured tool calls.

This guide covers deploying an MCP server from the web application, getting the connection URL, and connecting it to AI tools.

Prerequisites​

  • An active data cluster with at least one dataset containing processed entries.
  • Writer or Owner role in your organization.

Available MCP Servers​

Currently, the platform provides one MCP server type for deployment:

ServerDescription
mcp-dataclusterProvides 7 tools for searching, reading, and browsing your document collections. This is the primary integration point for AI agents.

See AI Agent Integration for the full list of tools and what they do.

Custom MCP Servers

We can build customized MCP servers tailored to your specific data sources and workflows. Contact us to discuss your needs.

Deploy from the Web Application​

Step 1: Navigate to the MCP Section​

  1. Click MCP in the left sidebar of the platform dashboard.
  2. The MCP page shows all your clusters with their current MCP deployment status.

Each row displays:

  • Cluster name and status (Active, Provisioning, etc.)
  • MCP status (Running, Not Deployed, Provisioning, Failed)
  • Created date

Step 2: Select an Image Tag​

For clusters that do not yet have an MCP deployed, an image tag dropdown appears in the row. Select the version you want to deploy (typically the latest available tag).

Step 3: Deploy​

  1. Click the Deploy button on the cluster row.
  2. The MCP status changes to Provisioning while the server is being set up on your data cluster.
  3. Once provisioning completes, the status changes to Running.

Deployment typically takes 1-2 minutes.

Step 4: Get the Connection URL​

Once the MCP server is running:

  1. Click the expand arrow on the left side of the cluster row.

  2. The connection card appears below, showing:

    • Server health status (Running / Healthy indicators)
    • Server URL — the MCP endpoint URL
    • claude_desktop_config.json — a ready-to-use configuration snippet for Claude Desktop

  3. Use the Copy buttons to copy the URL or the full configuration snippet.

Connect to Claude Desktop​

  1. Copy the claude_desktop_config.json snippet from the connection card.
  2. Open Claude Desktop and go to Settings > MCP Servers.
  3. Paste the configuration snippet.
  4. Claude will connect to your MCP server and its tools will become available.

The configuration snippet includes the server URL and authentication headers. It looks like this:

{
  "mcpServers": {
    "my-data-cluster": {
      "type": "http",
      "url": "https://your-cluster.mcp.alien.club/mcp",
      "headers": {
        "Authorization": "Bearer oat_YOUR_API_TOKEN"
      }
    }
  }
}

Once connected, you can ask Claude questions about your documents and it will use the MCP tools to search and read your data.

Connect to Claude Code​

For Claude Code, create a .mcp.json file in your project root with the same configuration:

{
  "mcpServers": {
    "my-data-cluster": {
      "type": "http",
      "url": "https://your-cluster.mcp.alien.club/mcp",
      "headers": {
        "Authorization": "Bearer oat_YOUR_API_TOKEN"
      }
    }
  }
}

Claude Code will automatically detect and connect to the MCP server.

Connect to Other AI Tools​

Any MCP-compatible AI tool can connect using:

  1. Server URL — Shown in the connection card after deployment.
  2. Authentication — Either OAuth (interactive) or API token header (programmatic).

The MCP server exposes standard discovery endpoints:

  • /.well-known/oauth-authorization-server — OAuth metadata for interactive authentication
  • /health — Health check endpoint for connectivity verification

See AI Agent Integration for details on authentication options.

Undeploy​

To remove an MCP server from a cluster:

  1. Navigate to the MCP section.
  2. Click Undeploy on the cluster row.

The MCP server is removed from the cluster. AI agents that were connected will lose access.

Troubleshooting​

MCP Status Shows "Failed"​

  • Check that your data cluster is in an Active state.
  • Try undeploying and redeploying the MCP server.
  • Contact us if the issue persists.

Health Shows "Unreachable"​

  • The MCP server may still be provisioning. Wait a minute and refresh the page.
  • Verify your data cluster is online and healthy.

AI Agent Cannot Connect​

  • Verify the server URL is correct (copy it fresh from the connection card).
  • Check that your API token is valid and has not expired.
  • Ensure the token belongs to the same organization that owns the cluster.

What's Next​

Your MCP server is deployed. Continue with: