kafka-connect-mcp

An MCP server that exposes Kafka Connect REST API operations as tools, letting LLMs manage connectors, tasks, and plugins through natural language.

Tools

Setup

Prerequisites

• Python 3.12+
• uv
• A running Kafka Connect cluster (or use the included Docker Compose)

Install

uv sync

Add to Claude Code

claude mcp add kafka-connect \
  -e KAFKA_CONNECT_URL=http://localhost:8083 \
  -- uv --directory /path/to/kafka-connect-mcp run kafka-connect-mcp

Configuration

Safe mode (capability-gated)

This server is read-only by default because all mutation capabilities default to false. Mutating tools are blocked unless you explicitly enable the specific capability.

Examples:

Restart-only mode for selected connectors:

KAFKA_CONNECT_ENABLE_RESTART=true
KAFKA_CONNECT_MUTATION_ALLOWLIST=payments-sink,inventory-source

Enable all mutating operations (for development only):

KAFKA_CONNECT_ENABLE_CREATE=true
KAFKA_CONNECT_ENABLE_UPDATE=true
KAFKA_CONNECT_ENABLE_DELETE=true
KAFKA_CONNECT_ENABLE_PAUSE_RESUME=true
KAFKA_CONNECT_ENABLE_RESTART=true

Running

stdio (default, for Claude Code)

KAFKA_CONNECT_URL=http://localhost:8083 uv run kafka-connect-mcp

SSE (for Docker / remote)

uv run kafka-connect-mcp --transport sse --host 0.0.0.0 --port 8000

Docker Compose (full stack)

Spins up Zookeeper, Kafka, Kafka Connect (with the Datagen plugin), and the MCP server:

docker compose up --build -d

Example: create a Datagen connector

Once the stack is up, ask Claude to create a datagen connector, or do it directly:

curl -X POST http://localhost:8083/connectors \
  -H "Content-Type: application/json" \
  -d '{
    "name": "datagen-users",
    "config": {
      "connector.class": "io.confluent.kafka.connect.datagen.DatagenConnector",
      "kafka.topic": "users",
      "quickstart": "users",
      "key.converter": "org.apache.kafka.connect.storage.StringConverter",
      "value.converter": "org.apache.kafka.connect.json.JsonConverter",
      "value.converter.schemas.enable": "false",
      "max.interval": "1000",
      "tasks.max": "1"
    }
  }'

Tests

uv run pytest tests/ -v

Tests use respx to mock HTTP calls to the Kafka Connect API -- no running cluster required.

Releases

Releases are automated from main:

1. Bump project.version in pyproject.toml (for example 0.1.0 -> 0.1.1).
2. Merge to main.
3. GitHub Actions creates tag/release vX.Y.Z.
4. The published release is automatically built and published to PyPI.

One-time PyPI setup

Set up PyPI Trusted Publishing for project kafka-connect-mcp:

• Owner: lawrencemq
• Repository: kafka-connect-mcp
• Workflow: .github/workflows/release.yml
• Environment: (none required by this workflow)

Project structure

kafka-connect-mcp/
├── pyproject.toml
├── Dockerfile
├── docker-compose.yml
├── src/kafka_connect_mcp/
│   ├── __init__.py
│   ├── safety.py            # Read-only and mutation policy gates
│   └── server.py            # MCP tools and entry point
└── tests/
    ├── conftest.py           # Shared fixtures (mock URL + respx router)
    ├── test_cluster.py
    ├── test_connectors.py
    ├── test_safety.py
    ├── test_tasks.py
    └── test_plugins.py

License

Licensed under the Apache License, Version 2.0. See LICENSE.