DegreeLume Assistant

The problem

Most California community-college students who set out to transfer never do: 19% transfer within four years, 28% within six (PPIC). Those who make it often carry courses that don't count; the median CSU transfer applicant arrives with 71.5 units against a 60-unit requirement. The answers exist, in ASSIST, the college catalog and Canvas, but in three systems that don't talk to each other, one campus pair at a time. This is built for the student with a job, the parent in evening classes, the first-generation student with no one to ask.

What it does

A college counselor an AI agent can call. The page registers 13 WebMCP tools with document.modelContext.registerTool():

• Transfer: check one course, audit a transcript, compare 17 UC and CSU campuses side by side (the question ASSIST cannot answer), explain any requirement. Every answer cites the agreement row and its year.
• School: the student's own Canvas, read-only: current grades, due work, a grade risk radar against the C-or-better rule, and every deadline in one list.
• Act: set the target, add reminders. The page shows what the agent did.

Why WebMCP fits

A deterministic engine over official records: structured in, cited out. A language model never decides; it only asks. Every call runs visibly and lands in an activity feed, so the student sees what was done and on what basis.

How it's built

React + TypeScript on Cloudflare Workers with a read-only Canvas proxy. The engine is DegreeLume's own pure functions over a public data slice: the El Camino College catalog and 52 ASSIST agreements. The Canvas token stays in the browser and is never stored. The demo runs on my real El Camino Canvas.

Sources: PPIC, Increasing Community College Transfers: Progress and Barriers (2020); PPIC, Improving Transfer from Community College to the CSU.

Built With

• canvas
• cloudflare
• react
• typescript
• vite
• webmcp