Functionize launches MCP server to link AI coding agents with live testing

Bridging the gap in AI coding

Functionize recently launched an MCP server designed to connect its testing platform with popular AI coding agents. These agents, which include GitHub Copilot, Cursor, and Claude Code, are increasingly used to generate and edit software code rapidly. However, Functionize notes a significant gap exists because these tools often lack direct visibility into how an application behaves when it is actually running.

While a coding agent can write a function quickly, it may not know if that change breaks a specific user flow in a live environment. Functionize argues that there is a risk when an agent that writes the code also creates the tests to verify it. In such cases, the tests might only confirm what the code says rather than whether the change matches the intended behavior of the software.

By providing an independent layer for software verification, Functionize aims to help engineering teams keep pace with the volume of code produced by generative AI. As the amount of newly generated code rises, the company suggests that tools for validating that code become more critical. This launch allows developers to stay inside their preferred tools instead of switching between separate coding and testing environments.

How the MCP integration works

The new server works with any client that supports the Model Context Protocol, which is a standard for connecting AI models to data sources. When a developer uses a compatible client, they can request a smoke or regression suite using plain language. The coding agent then passes information to Functionize Studio, including the requirements for the change, the specific code that was altered, and areas where test coverage is currently limited.

Functionize Studio takes this context and combines it with its own records of how the application behaves. It then traces the changes to affected user flows and tests those specific paths against the live application. The system returns the results to the developer in the same chat interface, providing details on status, step counts, and the exact step where a failure occurred.

To keep the process efficient, Studio uses task-specific models trained in-house for testing work. This approach is intended to be less expensive than using general-purpose AI models for the same tasks. Additionally, Studio updates its model of an application as pages change, which helps reduce the maintenance work of rewriting selectors when the software's interface is updated.

Targeting large-scale engineering

Functionize is positioning this product as an agentic quality platform focused on software verification. The Studio product handles the entire testing cycle, including authoring, execution, diagnostics, and reporting. Because of this comprehensive scope, the company is primarily targeting large corporate organizations that require automated testing tied closely to their release processes.

The company lists several large corporate customers as references for its platform, including GE HealthCare, McAfee, Markel, GEICO, and Alight. These names suggest that the tool is built for enterprise-level needs where reliability is paramount. The goal is to provide a system that can handle the scale of modern software development while maintaining high standards for quality.

The MCP server is currently available to existing Functionize Studio customers and those who are currently trialing the product. By making it compatible with various MCP-compliant clients, Functionize is attempting to fit into a wider ecosystem of AI tools. This strategy allows teams to adopt the new functionality without having to change their existing workflows or adopt a completely separate interface.

The role of human oversight

Despite the automation provided by the MCP server and the coding agents, Functionize emphasizes that human users remain responsible for the final decisions. The system is designed to provide the data needed to make informed choices, but it does not replace the human role in deciding what the software is meant to do. The human remains the ultimate authority on when a piece of software is ready to be released to the public.

The company describes the interaction as 'agent-to-agent' work that is made intelligent by shared context. By handing Studio the context behind every change, the coding agent can work more effectively with the testing platform. This creates a loop where the speed of code writing is matched by the speed of verification, but the direction of that work is still guided by human requirements.

As engineering teams look for ways to manage the explosion of AI-generated code, this distinction between generation and verification is becoming more relevant. Functionize aims to provide a way for teams to validate that the software functions as intended. This ensures that the rapid pace of development does not come at the expense of the user experience or the stability of the application.