Conduktor — Claude Code MCP server: what it does, how to install
Connects AI assistants to Apache Kafka via Conduktor Console to inspect clusters, consumer groups, schemas, and governance rules.
claude mcp add kafka-mcp -e CONDUKTOR_CONSOLE_URL=<CONDUKTOR_CONSOLE_URL> -e CONDUKTOR_API_TOKEN=<CONDUKTOR_API_TOKEN> -- npx -y @conduktor/mcpWhat Conduktor does
The server turns an AI assistant into an operator for Apache Kafka infrastructure by connecting to Conduktor Console. Instead of streaming raw message logs to the model, it exposes control plane metadata covering topics, configurations, and connectors. Calls execute under the caller's console identity. This structure enforces existing access rules and logs every action.
Built-in actions let agents inspect broker health, track lagging consumer groups, verify schema evolution against the registry, and trace stream lineage. When debugging a stuck pipeline, the model can look up specific records by offset or check which certificates approach expiration.
Running this integration requires an active Conduktor Console deployment alongside a Personal Access Token. The npm launcher @conduktor/mcp acts as a client proxy to the console endpoint, meaning cluster records stay inside Kafka unless explicitly requested.
What it helps with
- detect consumer group lag and identify stuck offsets
- verify schema compatibility before publishing changes
- locate abandoned topics and assign ownership tags
- monitor certificate expirations across Kafka clusters
Example prompts for the agent
- Show consumer group lag across the cluster and find the record blocking the stuck consumer.
- List topics without recent consumer activity and inspect their creator details.
- Check whether the updated schema is compatible with the current version in Schema Registry.
How to install Conduktor
1Platform CLI
claude mcp add kafka-mcp -e CONDUKTOR_CONSOLE_URL=<CONDUKTOR_CONSOLE_URL> -e CONDUKTOR_API_TOKEN=<CONDUKTOR_API_TOKEN> -- npx -y @conduktor/mcpReplace <…> with your own values
2Platform CLI: remote server
claude mcp add --transport http conduktor https://{console_host}/api/mcp1Platform CLI
codex mcp add conduktor --env CONDUKTOR_CONSOLE_URL=<CONDUKTOR_CONSOLE_URL> --env CONDUKTOR_API_TOKEN=<CONDUKTOR_API_TOKEN> -- npx -y @conduktor/mcpReplace <…> with your own values
2Platform CLI: remote server
codex mcp add conduktor --url https://{console_host}/api/mcp1Config file
{
"mcpServers": {
"conduktor": {
"command": "npx",
"args": [
"-y",
"@conduktor/mcp"
],
"env": {
"CONDUKTOR_CONSOLE_URL": "<CONDUKTOR_CONSOLE_URL>",
"CONDUKTOR_API_TOKEN": "<CONDUKTOR_API_TOKEN>"
}
}
}
}Replace <…> with your own values
2Config file
{
"mcpServers": {
"conduktor": {
"url": "https://{console_host}/api/mcp"
}
}
}1Config file
{
"servers": {
"conduktor": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@conduktor/mcp"
],
"env": {
"CONDUKTOR_CONSOLE_URL": "<CONDUKTOR_CONSOLE_URL>",
"CONDUKTOR_API_TOKEN": "<CONDUKTOR_API_TOKEN>"
}
}
}
}Replace <…> with your own values
2Config file
{
"servers": {
"conduktor": {
"type": "http",
"url": "https://{console_host}/api/mcp"
}
}
}1Platform CLI
gemini mcp add -e CONDUKTOR_CONSOLE_URL=<CONDUKTOR_CONSOLE_URL> -e CONDUKTOR_API_TOKEN=<CONDUKTOR_API_TOKEN> conduktor npx -y @conduktor/mcpReplace <…> with your own values
2Platform CLI: remote server
gemini mcp add -t http conduktor https://{console_host}/api/mcpStep by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about Conduktor
How to install Conduktor in Claude Code?
Run claude mcp add kafka-mcp -e CONDUKTOR_CONSOLE_URL=<CONDUKTOR_CONSOLE_URL> -e CONDUKTOR_API_TOKEN=<CONDUKTOR_API_TOKEN> -- npx -y @conduktor/mcp with your console URL and access token.
What does the Conduktor MCP server do?
It enables AI agents to query and manage Kafka resources through Conduktor Console, handling topics, schemas, consumer groups, and access controls.
Does Conduktor expose message payloads to the AI model?
The integration focuses on control plane metadata, so message payloads stay within Kafka unless an assistant invokes a tool designed to read specific partition offsets.
