New functions for AI, network and Java projects with Docker Desktop 4.2

Version 4.42 of Docker Desktop introduces new network rules and integrates the MCP toolkit for easy access to AI sources via the MCP protocol (Model Context Protocol). The Model Runner improves the handling of AI models and their deployment.

Docker Desktop, the Docker tool for Windows and Mac, now supports IPv6 natively. Users can choose whether they want to run either IPv4 or IPv6 or both in parallel. Docker Desktop recognizes the host's network stack during DNS resolution and filters out unsupported record types. This is intended to prevent timeouts if users have only selected one IP version. The corresponding settings can be found under Settings/Resources/Network section and admins can specify them system-wide.

AI sources and model under control

Docker Desktop has now fully integrated the MCP toolkit introduced in April, allowing users to quickly integrate MCP sources from the Docker catalog into their systems. There are now over a hundred images, for example from GitHub, MongoDB or HashiCorp, which can be linked to Claude Desktop, Cursor, Continue.dev or Docker's AI agent Gordon. All MCP images are signed and run in a shielded environment. They can be managed in the console with the new mcp command.

There are also new features for the Docker Model Runner introduced in Docker Desktop 4.40, which is based internally on the latest version of llama.cpp and now also supports Windows devices with Qualcomm chips. The Model Runner also runs on the Docker Engine Community Edition across many Linux distributions. According to the blog entry, this is helpful for developers who want to integrate AI capabilities into their pipelines and test runs. There are also extended console functions for this, for example for packing OCI-compatible formats.

The Dockerize function has also been updated, which now simplifies the handling of Java, Kotlin, Gradle and Maven projects by minimizing the configuration effort.

(who)