Confluent (Kafka) MCP Server

The Confluent MCP server is an official open-source Model Context Protocol server maintained by Confluent that lets AI assistants interact with Confluent Cloud, Confluent Platform, and Apache Kafka deployments through natural language. It exposes more than 50 tools spanning Kafka topic and message operations, Flink SQL statement management, Schema Registry, Connectors, Tableflow, metrics and billing queries, tags, and catalog search.

The server supports three transport mechanisms (stdio, HTTP, and SSE) and two authentication modes: static API keys configured in a YAML file, or OAuth (PKCE) browser-based sign-in via Confluent Cloud for supported tool categories. It also includes always-available documentation tools (search-product-docs, get-product-doc-page) and diagnostics (explain-disabled-tools) so agents can discover which capabilities are usable in the current configuration.

Because it is maintained by Confluent itself and distributed as the @confluentinc/mcp-confluent npm package, it is the canonical way to plug Kafka and Flink workflows into Claude Desktop, Cursor, VS Code, Goose, Gemini CLI, and Windsurf.

Tools

Prerequisites

• Node.js 22 or higher
• A Confluent Cloud account (for cloud features) or a local Kafka/Schema Registry endpoint
• Confluent Cloud API keys, or use OAuth (PKCE) browser sign-in

Install and initialize

# Generate a starter config.yaml
npx @confluentinc/mcp-confluent --init-config

# Edit config.yaml with your bootstrap servers, API keys, endpoints, etc.

# (Optional) Initialize OAuth instead of static keys
npx @confluentinc/mcp-confluent --init-oauth-config

Example minimal config.yaml:

connections:
  local:
    type: direct
    kafka:
      bootstrap_servers: "localhost:9092"
    schema_registry:
      endpoint: "http://localhost:8081"

Claude Desktop config

Add the following to claude_desktop_config.json (paths from the repo's example.claude_desktop_config.json):

{
  "mcpServers": {
    "confluent": {
      "command": "node",
      "args": [
        "/path/to/confluent-mcp-server/dist/index.js",
        "--config",
        "/path/to/confluent-mcp-server/config.yaml"
      ]
    }
  }
}

You can also run via npx directly:

{
  "mcpServers": {
    "confluent": {
      "command": "npx",
      "args": ["-y", "@confluentinc/mcp-confluent", "--config", "/absolute/path/to/config.yaml"]
    }
  }
}

Transports

The server supports stdio (default), http, and sse. Enable multiple with:

npx @confluentinc/mcp-confluent --transport http,sse,stdio

When exposing HTTP/SSE, set server.auth.api_key in config.yaml and configure server.http.port.

Useful flags

• --list-tools: print all tools available for the current configuration
• DO_NOT_TRACK=true: disable telemetry

• Create, list, and reconfigure Kafka topics across Confluent Cloud or local clusters from a chat interface.
• Produce test messages and consume from topics to validate pipelines without writing client code.
• Author, submit, and inspect Flink SQL statements on Confluent Cloud compute pools.
• Manage Kafka Connect connectors (create, update, list) and troubleshoot Tableflow integrations.
• Search Confluent product documentation and diagnose disabled tools while building agents.

• "List all topics in my Confluent Cloud cluster and show partition counts."
• "Create a new topic called orders-events with 6 partitions and replication factor 3."
• "Submit this Flink SQL statement to compute pool lfcp-prod and report any errors."
• "Show me the last 20 messages on payments.failed and summarize the error codes."
• "Find docs explaining how to set up Tableflow with Iceberg catalogs."

• Officially maintained by Confluent and published to npm as @confluentinc/mcp-confluent.
• Broad coverage: 50+ tools across Kafka, Flink SQL, Schema Registry, Connectors, and Tableflow.
• Supports multiple transports (stdio, HTTP, SSE) and both static API keys and OAuth (PKCE).
• Works with Claude Desktop, Claude Code, Cursor, VS Code, Goose, Gemini CLI, and Windsurf.

• Requires Node.js 22+ and a YAML configuration step before first use.
• Some tool categories still require static API keys; OAuth coverage is partial and migration is ongoing.
• Tableflow operations require prerequisite IAM role setup in Confluent Cloud.

• kafka-mcp-server (community) community implementations targeting plain Apache Kafka clusters without Confluent Cloud features.
• Direct use of the Confluent CLI (confluent) wrapped via a generic shell MCP server for users who prefer CLI semantics.
• Redpanda's own tooling for teams running Redpanda instead of Confluent/Kafka.