Functionize has released an MCP server that links its testing platform with coding agents including Claude Code, Cursor and GitHub Copilot. The launch is intended to let developers and testers run checks against a live application from the same chat interface.
The product connects Functionize Studio to MCP-compatible clients, allowing users to request a smoke or regression suite in plain language and receive the result in the conversation. Studio builds the tests, runs them against the application, and returns details on what failed and why.
The move is aimed at a gap in the growing use of coding agents in software development. While those tools can generate and edit code quickly, Functionize argues they do not have direct visibility into how an application behaves when running, or whether a change affects user flows as intended.
In this setup, the coding agent passes Studio information including the requirement behind a change, the code that was altered, and areas where test coverage is limited. Studio combines that with its own record of application behaviour, traces the change to affected user flows, and tests those flows against the live application.
The result is returned to the developer in the same interface, including status information, step counts, and the first failing step. Functionize says this lets engineering teams stay inside tools they already use instead of moving between separate coding and testing environments.
Independent checks
Functionize is positioning the product as an independent layer for software verification at a time when more code is being produced by AI systems. It argues that tests created by the same agent that wrote the code may confirm what the code says rather than whether the change matches the application's intended behaviour.
That distinction is becoming more relevant as engineering teams look for ways to keep pace with code produced by generative AI tools. Functionize says the volume of newly generated code is rising faster than many teams can review and test through existing workflows.
"As the volume of newly generated code explodes, the tools we use to validate it become critical," said Tamas Cser, Founder and Chief Executive Officer of Functionize. "Studio has spent years learning how applications behave. With MCP, the coding agent hands Studio the context behind every change. Studio verifies the affected user flows against the application and pinpoints the exact step where something broke. That shared context is what makes agent-to-agent work intelligent. Developers drive Studio from the tools they already use, and every change is checked against how the app is supposed to work, at the speed the code is being written."
How it works
Functionize says the MCP server works with Claude Code, Claude Desktop, Cursor and GitHub Copilot in Visual Studio Code, as well as other clients that support the Model Context Protocol. A Claude Code plugin adds what it describes as extra skills to help the coding agent work with Studio.
Studio also updates its model of an application as pages change, allowing tests to keep running without developers having to rewrite selectors. That is intended to reduce maintenance work, a long-standing challenge in automated software testing.
Another element of the launch is cost. Functionize says Studio runs on task-specific models trained in-house for testing work, reducing the expense of building and running tests compared with using general-purpose models for the same process.
Broader push
Functionize describes itself as an agentic quality platform focused on software verification. Its Studio product covers authoring, execution, maintenance, diagnostics and reporting across the testing cycle.
The company says the system has been trained on nine years of execution data and is used by large corporate customers including GE HealthCare, McAfee, Markel, GEICO and Alight. Those references suggest Functionize is targeting larger engineering organisations that need automated testing tied closely to release processes.
The MCP server is available to existing Functionize Studio customers and trial users. It is designed to work with any MCP-compatible client, reflecting the company's effort to fit into a wider ecosystem of AI coding tools rather than requiring teams to adopt a separate interface.
The central claim behind the launch is that coding agents can generate changes quickly but still need a separate system to check whether those changes work in the application itself, with the human user remaining responsible for deciding what the software is meant to do and when it is ready to be released.