On a bare config you get documentation and diagnostics; everything else appears per connection once the config it needs is present. That inversion turns the usual guessing game into a lookup — one tool takes a tool name and returns the YAML that would unlock it, another groups every gap by the piece of config it waits on, and neither writes to your file. If you are a Confluent Cloud customer and want none of this, Confluent also runs a managed server; this one is the answer for Confluent Platform, self-managed Kafka, or a toolset you intend to trim yourself.
Confluent's open-source server for Confluent Cloud, Confluent Platform and standalone Kafka. Documentation and diagnostics are always present; the Kafka, Flink, Schema Registry, Connector, Tableflow, Metrics and Billing tools switch on per connection as the configuration for them appears.
- Kafka worked from the client: topics listed, created, deleted and reconfigured, messages produced and consumed, consumer groups described with their lag
- Flink SQL statements created, listed, deleted and read back, with catalogs, databases and table schemas explored, plus health checks, issue detection and a query profile
- Connectors inspected and managed — status, tasks, offsets, logs, an error summary and suggested remediation, then pause, resume, restart or reconfigure
- Schema Registry subjects listed, registered and deleted, and topics organised and searched by tag
- Tableflow topics and catalog integrations created and managed, alongside Confluent Cloud metrics and billing costs
- Keyword search across docs.confluent.io, developer.confluent.io and support.confluent.io, then the full markdown of any result page — available on a bare config with no credentials
- Diagnostics that explain the tool list itself: every configured connection with the tools routable against it and whether it is read-only, the unavailable ones bucketed by the config each is waiting on, and a copy-pasteable YAML snippet for the tool you want to unlock
Node.js 22.19.0 or later, and a YAML configuration — the CLI writes a starter one for you. Confluent Cloud takes either static API keys per service block or an OAuth sign-in in the browser, which covers most tool groups but not all of them yet; a local Kafka or Confluent Platform needs only bootstrap servers and a Schema Registry endpoint. Which tools appear follows the blocks you fill in, and allow and block lists trim the surface further. Over HTTP or SSE the server wants an API key of its own, generated by the CLI. Tableflow additionally needs IAM permissions granted in your cloud account before any of its calls succeed. Usage data and error reports are sent unless DO_NOT_TRACK=true. MIT, community-supported.
One command plus a key — npx @confluentinc/mcp-confluent --config ./config.yaml, then supply credentials