I´ve noticed that the future (now really) is having many IA agents running at same time on different projects, wich already causes an overhead, and i´ve realized that maybe an MCP for codeception could be useful. I´was thinking about exposing the same methods as a FunctionalTester does, so AI could self test applications without having to write files. What do you think about?

Proposal: MCP for on-the-fly functional testing with Codeception

This proposal suggests creating an MCP (Model Context Protocol) that exposes capabilities similar to Codeception’s FunctionalTester, but as an endpoint.

Overview

The MCP would bootstrap the application directly in memory, just like functional tests already do, but without relying on a browser or even a running web server.
It would provide a set of methods equivalent to FunctionalTester (e.g., sendGET, sendPOST, seeResponseCodeIs, seeResponseContainsJson, etc.) accessible via MCP calls.
This would allow executing requests, inspecting responses, and performing assertions dynamically.

Benefits

• Low overhead: since execution happens entirely in memory, there’s no need for a full browser or HTTP stack.
• AI integration: an AI could leverage this MCP to dynamically test application behavior, validate expected outputs, and—once confirmed—automatically generate the final test to be added to the suite.
• Reusing existing infrastructure: it builds on Codeception’s functional testing layer, ensuring compatibility with existing workflows.

Potential applications

• Validating new features on the fly during development.
• Generating tests automatically based on real usage scenarios.
• AI-assisted debugging and verification without heavy browser-based setups.