Connect an AI assistant to Conduktor MCP

Connect Claude Desktop or another MCP client to Conduktor Console with a personal access token, and ask questions about your Kafka platform.

Connect an AI assistant to the Model Context Protocol (MCP) server built into Conduktor Console. The assistant then uses your own Console permissions to answer questions about your Kafka platform.

Prerequisites

Requirement	Required	Details
MCP server enabled in Console	Yes	Enabled by default. Admins can turn it off with enable_mcp: false (CDK_ENABLE_MCP=false).
MCP client	Yes	Claude Desktop, or any client that can connect to a remote MCP server.
Node.js and npm	Yes, for Claude Desktop	The Claude Desktop configuration uses npx to run mcp-remote, which bridges it to Console. Download it from nodejs.org.
Insights enabled	No	Needed for the insights-cluster, insights-topics, query-topics and aggregate-topics tools. See Insights.

Set up the connection

1. Open the MCP setup page

   In Console, go to Settings > MCP.

   
2. Generate a token

   Click Generate token and configure. Console creates a personal access token named MCP Token - <date> and adds it to the configuration shown on the page.

   Copy the configuration straight away: the token won’t be visible once you leave the page.

   To create the token yourself, click create a token manually or go to Settings > API Keys. Find out more about personal access tokens.

3. Add the configuration to your MCP client

   Claude Desktop

   In Claude Desktop, go to Settings > Developer and click Edit Config to open claude_desktop_config.json:

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

   Paste the configuration from Console under mcpServers. It looks like this:

   {
     "mcpServers": {
       "conduktor-console": {
         "command": "npx",
         "args": [
           "mcp-remote",
           "<CONSOLE_URL>/api/mcp",
           "--header",
           "Authorization: Bearer ${CONDUKTOR_API_TOKEN}"
         ],
         "env": {
           "CONDUKTOR_API_TOKEN": "<YOUR_PERSONAL_ACCESS_TOKEN>"
         }
       }
     }
   }
   

   Other MCP clients

   Configure a remote MCP server with these details:

   Setting	Value
   Endpoint	<CONSOLE_URL>/api/mcp, for example https://console.example.com/api/mcp
   Header	Authorization: Bearer <YOUR_PERSONAL_ACCESS_TOKEN>

   If your client can’t send a custom header to a remote server, run it through mcp-remote with the same command and args as the Claude Desktop configuration.

4. Restart your client and test the connection

   Restart your MCP client, then ask it a question such as:

   Which Kafka clusters can you see in Conduktor?
   

   The assistant calls list-clusters and lists the clusters you have access to in Console. If it does, you’re connected.

Manage access

The token you connect with decides what the assistant can see and do:

• Personal access token: inherits your own permissions in Console. Use this for MCP clients.
• Admin API key: also works, but has admin access to every resource. Avoid it for AI assistants.

To disconnect an assistant, revoke its token in Settings > API Keys. Find out more about revoking API keys.

Troubleshoot the connection

The assistant doesn’t list some topics, consumer groups or other resources: tools that list resources leave out the ones your token can’t access, without reporting an error. Check your permissions in Console.

The Insights tools, query-topics or aggregate-topics are missing: these tools are only available when Insights is enabled. Ask your admin to check the enable_insights setting. See Disable Insights.

The client can’t connect or gets an authentication error: check that the endpoint ends in /api/mcp and that the token hasn’t been revoked. If you use Claude Desktop, check that Node.js and npm are installed and that npx is on your PATH.

Next steps

Find out which tools the assistant can use, and what you can ask it about your topics with aggregate-topics.