Creativity · MCP — server

MCP Kafka Server

Apache Kafka is the dominant distributed log for event streaming. The MCP Kafka Server lets Claude or another MCP client list topics, inspect consumer group lag, and publish or read messages through tool calls — a handy way to debug production streaming pipelines from a chat interface.

MCP facts

Kind: server
Ecosystem: anthropic-mcp
Language: Python
Transports: stdio

Capabilities

Tools: list_topics, describe_topic, consumer_lag, produce_message, consume_messages
Authentication via SASL/PLAIN, SASL/SCRAM, or mTLS
Works with Confluent Cloud, MSK, Redpanda

Install

pip install mcp-kafka

Configuration

{
  "mcpServers": {
    "kafka": {
      "command": "uvx",
      "args": ["mcp-kafka"],
      "env": {
        "KAFKA_BOOTSTRAP_SERVERS": "pkc-xxx.confluent.cloud:9092",
        "KAFKA_SASL_USERNAME": "<key>",
        "KAFKA_SASL_PASSWORD": "<secret>",
        "KAFKA_SECURITY_PROTOCOL": "SASL_SSL"
      }
    }
  }
}

Frequently asked questions

Can the model produce messages?

Yes, but gate it behind explicit approval — publishing to a production topic from a chat is powerful and easy to misuse. Many deployments limit this to staging brokers.

Does it work with Redpanda?

Yes — Redpanda is Kafka-API compatible, so the same server works unchanged.

Good use cases?

Checking consumer lag during incidents, peeking at recent events on a topic to debug producers, and producing test messages to staging.

Sources

Apache Kafka Documentation — accessed 2026-04-20
Model Context Protocol — accessed 2026-04-20