Kafka MCP Server

The Kafka MCP Server is a Model Context Protocol (MCP) server that provides integration with Apache Kafka to enable interaction with a Kafka cluster using natural language (LLMs).

demo

Prerequisites

You will either need Docker or Golang to run the MCP server locally.

Getting Started

You need to have access to a Kafka cluster. Following the quickstart doc:

 docker pull apache/kafka:4.0.0
 docker run -p 9092:9092 apache/kafka:4.0.0

Kafka is available on localhost:9092 now.

Usage with Claude Desktop

Docker:

{
  "mcpServers": {
    "kafka": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "KAFKA_MCP_BOOTSTRAP_SERVERS",
        "ghcr.io/cefboud/kafka-mcp-server"
      ],
      "env": {
        "KAFKA_MCP_BOOTSTRAP_SERVERS": "localhost:9092"
      }
    }
  }
}

Building locally

cd <workdir>
git clone https://github.com/CefBoud/kafka-mcp-server.git
cd kafka-mcp-server
go build -o kafka-mcp-server  cmd/kafka-mcp-server/main.go 

{
  "mcpServers": {
    "kafka": {
      "command": "<workdir>/kafka-mcp-server/kafka-mcp-server",
      "args": [
        "stdio",
      ],
      "env": {
        "KAFKA_MCP_BOOTSTRAP_SERVERS": "localhost:9092"
      }
    }
  }
}

Options:

      --bootstrap-servers string   Comma-separated list of the Kafka servers to connect to.
      --enable-command-logging     When enabled, the server will log all command requests and responses to the log file
      --log-file string            Path to log file
      --read-only                  Restrict the server to read-only operations

All options can be passed as environment variables, uppercased, with hyphens replaced by underscores, and prefixed with MCP_KAFKA_ e.g., --bootstrap-servers becomes MCP_KAFKA_BOOTSTRAP_SERVERS.

Available MCP Tools

• List topics
• Create topic
• Consuming messages.
• Produce messages.
• Describe the clusters (list of brokers and controller)
• List consumer groups and their lag.
• Get topic's earliest and latest offsets (GetOffsetShell)
• Reset consumer group offsets.
• Kafka Connect ??
• Schema Registry ??

🔀 MultiplexTool

demo MultiplexTool

Running many sequential tools, especially when each depends on the output of the previous, can be tedious and time-consuming. This requires multiple round-trips between the client and server/

MultiplexTool solves this by allowing the client to batch a list of tool calls into a single request, executing them in order. It supports dynamic dependencies between tools by letting you reference earlier outputs using prompt-based placeholders.

If a tool input depends on a previous result, the client uses the PROMPT_ARGUMENT: format to generate that input dynamically via a prompt to an LLM (currently only Gemini is supported).
Example:
"userId": "PROMPT_ARGUMENT: the ID of the created user"

CLI Flags:

• --enable-multiplex: Enables multiplexing of tool calls.
• --multiplex-model: Specifies the model (e.g. gemini) used to infer PROMPT_ARGUMENTs. Requires GEMINI_API_KEY env variable.

{
  "mcpServers": {
    "kafka": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "KAFKA_MCP_BOOTSTRAP_SERVERS",
        "ghcr.io/cefboud/kafka-mcp-server",
        "--enable-multiplex",
        "--multiplex-model",
        "gemini"
      ],
      "env": {
        "KAFKA_MCP_BOOTSTRAP_SERVERS": "localhost:9092",
        "GEMINI_API_KEY": "....."
      }
    }
  }
}

Credits

• github.com/github/github-mcp-server
• github.com/mark3labs/mcp-go
• github.com/IBM/sarama