JetBrains adds MCP server to let AI agents verify Compose Multiplatform UI

JetBrains has introduced a new capability that allows AI coding agents to independently verify changes they make to Compose Multiplatform applications. This integration creates a functional feedback loop where agents can validate their own code by interacting with a live version of the software.

As part of the Compose Multiplatform 1.12.0 release, JetBrains implemented an experimental Model Context Protocol (MCP) server within Compose Hot Reload. This server allows AI agents to connect directly to running applications to ensure that code modifications produce the intended results. According to technical documentation, agents can now trigger reloads, capture screenshots, and inspect the semantic structure of the user interface. Furthermore, the agents can read application logs and perform active UI interactions, such as clicking elements and entering text.

The shift toward visual feedback

Compose Multiplatform extends Android's modern UI toolkit, bringing the declarative framework to desktop and web environments. Traditionally, AI coding assistants have relied on static analysis—predicting how code will behave based on patterns in the text. By integrating with a live runtime via MCP, JetBrains is moving toward autonomous agents that operate based on visual and structural feedback. This allows the agent to "see" the result of a change and iterate on the code until the UI matches the desired specification, rather than relying on a human developer to spot a visual glitch.

Reducing the manual review cycle

The ability for AI agents to autonomously verify UI changes represents a significant shift in the mobile and cross-platform development workflow. By handling the initial verification phase, these agents can substantially reduce the manual review cycle for developers. This automation decreases the likelihood of visual regressions—where a change in one part of the app inadvertently breaks the layout elsewhere—because the agent can programmatically check the interface before the code ever reaches a human reviewer.

Future of autonomous UI development

As this experimental MCP server matures, the industry will be watching to see how deeply these agents can integrate into the full CI/CD pipeline. While the current capability focuses on the development phase via Hot Reload, the potential for agents to perform autonomous end-to-end testing on multiplatform apps is a logical next step. It remains to be seen how JetBrains will refine the security and stability of these agent-to-app connections as the feature moves beyond the experimental stage.