Foray - an MCP-governed agent squad that turns student confusion into faculty verified corrections

About this project

What it does: A student speaks a short reflection on what they just learned. Their speech is transcribed into text, then one agent identifies any misconceptions using a fixed taxonomy. A second agent searches the actual course material for the correct explanation and verifies it with the relevant page before sending it back. If it cannot find a grounded citation, it does not generate a correction. Instead, the case is escalated to the faculty. Across an entire class, deterministic code groups similar misconceptions into clusters, and another agent traces each cluster through a prerequisite graph to identify the underlying concept students are struggling with. A week later, the system checks the same misconception again and reports whether it has been resolved. Who it's for: Primarily faculty, who often discover what students misunderstood only weeks later through exam scores that reveal little about the reason behind the mistakes. It also benefits students by creating a learning roadmap based on demonstrated mastery rather than self-reported interests. What makes it special: Speaking is more natural and requires less effort than writing. Students are often more honest and expressive when they explain what confused them out loud, making speech transcription an essential part of capturing meaningful learning signals. Each agent runs behind its own MCP server with access only to the tools required for its specific role, so the separation between a student's raw voice reflections and the faculty's aggregated insights is enforced through system permissions rather than policy alone. Every correction must be backed by evidence from the course material before it reaches the student, preventing unsupported responses. Finally, the system closes the feedback loop by checking a week later whether the misconception has actually been corrected instead of assuming the intervention was successful.