Confluent MCP Server is an open-source MCP server that enables AI assistants to interact with Confluent Cloud, Confluent Platform, and standalone Apache Kafka deployments through natural language. It provides more than 50 tools covering Kafka, Flink SQL, Schema Registry, Connectors, Tableflow, and more, and can be used with any MCP-compatible client, including Claude Desktop, Claude Code, Cursor, VS Code, Goose, and Gemini CLI.
Purpose Confluent MCP Server is used to connect AI assistants to resources in the Confluent ecosystem: AI clients can use natural language to invoke the tools it provides and interact with Confluent Cloud, Confluent Platform, or local Kafka / Schema Registry deployments. It is suitable for scenarios where Confluent Cloud is already in use and users want AI-assisted operation of resources, as well as for users who need Confluent Platform / self-managed Kafka support or want to customize and extend the toolset.
Main capabilities Tools are automatically enabled according to the service blocks enabled in the configuration file. Tools that are always available require no service block or authentication and are available even if the configuration is empty. Tools for Confluent Cloud require endpoints and authentication for specific Confluent Cloud components, and some categories support OAuth authentication (login through the browser rather than configuring API keys). Tools for local deployments require only a Kafka or Schema Registry endpoint and do not require a Confluent Cloud API key/secret pair, making them suitable for local development using self-managed clusters (including Confluent Platform). Confluent Cloud-specific tools (such as Flink, Tableflow, Billing, Metrics, etc.) require a Confluent Cloud account and remain disabled on Confluent Platform.
Usage Prerequisites: Node.js 22.19.0 or later. If you need to interact with Confluent Cloud, you must first create an account. First generate a config.yaml file in the project root directory and fill in the connection information. There are two ways to start the server: build and run from source (install dependencies with pnpm install, build the project with pnpm run build or pnpm run dev), or use npx directly (no build required): npx @confluentinc/mcp-confluent --config /path/to/myconfig.yaml. Then configure the MCP client to connect to the address where the server is located (usually localhost on a specific port, with the port set by server.http.port in config.yaml). After starting the MCP client, the client will run an instance of this MCP server locally at startup, which is responsible for managing data schemas and interacting with resources on behalf of the user. You can interact with Confluent Cloud or local resources through the client interface, and the client will send requests to this MCP server, which will interact with the available connections on its behalf.
Authentication and configuration In addition to supporting static API keys defined in the YAML configuration, the server also supports authenticating to Confluent Cloud through OAuth (PKCE): on the first tool call that requires Confluent access, the server opens a browser and redirects to the Confluent Cloud login page, and subsequent tool calls reuse that session without configuring API keys. You can use --init-oauth-config to generate a starting configuration. YAML (-c config.yaml) and the legacy environment variable path (-e config.env) are fully compatible for a single connection; the environment-variable-only path will emit a startup warning in future versions and will eventually be removed. Multiple connections (or no connections) are supported only in YAML.
Tableflow prerequisites Tableflow tools interact with cloud storage (such as AWS S3) and metadata catalogs (such as AWS Glue) on behalf of the user through the Confluent Cloud Flink runtime. The Flink runtime requires IAM permissions for the cloud account and must be granted and linked to Confluent Cloud before any Tableflow tool can succeed. Skipping this step will cause mcp-confluent to attempt to configure or manage TaAn authorization error occurs when viewing the bleflow table.
Use cases Suitable for users who want to operate Confluent Cloud, Confluent Platform, or local Kafka / Schema Registry through an AI assistant using natural language, especially suitable for customers already using Confluent Cloud, as well as developers who need self-managed Kafka support or want to customize and extend the toolset. The connection method for a Confluent Platform cluster is the same as any local Kafka + Schema Registry deployment: point to the broker and Schema Registry; the only differences from the localhost:9092 setup are authentication and TLS.
Limitations This open source MCP server is a community-supported project. Confluent does not provide dedicated support, only best-effort support, with no service-level commitment. If you encounter problems or want to contribute, you can submit an issue or pull request directly in this repository. The Node.js version must be 22.19.0 or higher. Tools require a specific service block in config.yaml to appear; use --list-tools to see the activated tools, or call the explain-disabled-tools tool to learn why an individual tool is disabled. HTTP/SSE authentication errors require generating an API key via npx @confluentinc/mcp-confluent --generate-key and adding it to server.auth.api_key. The default HTTP port is 8080 and can be changed via server.http.port. This server collects usage data and reports server-side runtime errors via Sentry; you can opt out by setting DO_NOT_TRACK=true or server.do_not_track: true. OAuth currently supports only the marked categories; the rest still require a static API key for direct connection.
