Omnivox MCP

Your own personal bridge to Omnivox — the Quebec college student portal — for AI assistants, side projects, and curious tinkerers.

This started life as an MCP server and grew into something broader: a self-hosted toolbox that logs into Omnivox once and then lets anything talk to it — Claude over MCP, any assistant that can call plain HTTP tools, your own scripts, or you with curl at 2am wondering if a grade got posted.

Under the hood it runs a persistent Puppeteer browser logged into your Omnivox session, executing requests through the site's own JavaScript — so challenge-response auth, cookies, and encoding are all handled natively by Omnivox's own code. On top of that browser it offers four things:

• An MCP server — 34 tools covering courses, grades (down to individual evaluations), schedule, calendar, MIO messaging, documents, assignments, and college news.

• A REST tool gateway — every MCP tool doubled as a plain HTTP endpoint, with a live catalog at GET /tools and a full OpenAPI 3.1 spec at GET /openapi.json. Virtually any assistant or agent framework that supports JSON-described tools can use it, MCP support or not. It's also just a nice API for your own projects.

• A raw Omnivox proxy — /Mobl/* passes any request straight through the authenticated session, for endpoints no tool wraps yet.

• A research effort — the mobile API this all sits on is undocumented, so we document it ourselves as we reverse-engineer it. The growing field notes live in docs/.

Everything runs on your machine (or your server, or a container). One instance = one Omnivox account.

Quick start

# 1. Clone and install
git clone https://github.com/Beat-YT/omnivox-mcp.git
cd omnivox-mcp
npm install

# 2. Authenticate (first time only)
cd omnivox-connection
npm install && npm start
# Log in through the Electron window, save cookies.json + config.json,
# then place them in the data/ folder at the project root (or your OMNIVOX_DATA_DIR)

# 3. Start the server
cd ..
npm start              # Express server on port 3000

The server starts with MCP-over-HTTP at /mcp?key=..., the REST tool gateway, and the proxy. An access key is auto-generated at data/accessKey.txt — that's your password to the whole thing, treat it accordingly.

(There's also a stdio mode, npm run start:stdio, where an MCP client launches the server as a subprocess — deprecated and not recommended.)

Docker

Prefer containers? Authenticate once with the Electron app, then copy the example compose file and set your data folder path:

cp docker-compose.example.yml docker-compose.yml
# Edit docker-compose.yml — replace /path/to/config/folder with your data folder
docker compose up -d

The included docker-compose.example.yml mounts your data folder into the container at /data/omnivox, exposes port 3000, sets a session-keeping refresh interval, and wires up a healthcheck. Your customized docker-compose.yml is gitignored.

Re-authentication

If your session expires or you want a different account, reset and log in again:

npm run reset
cd omnivox-connection && npm start
# Log in, save the two files, place them back in your data folder
cd .. && npm start

npm run reset clears browser/, cookies.json, config.json, and accessKey.txt from the data folder so the server starts fresh.

Configuration

Set these as environment variables or in a .env file at the project root:

Related MCP server: Canvas MCP Server

Using it

With an MCP assistant

Connect your MCP client to http://localhost:3000/mcp?key=YOUR_KEY via Streamable HTTP transport. That's it — the assistant gets all 34 tools and you can ask about grades, schedules, assignments, or messages in natural language. See AGENT_SETUP.md for client configs.

With any other assistant (REST tool gateway)

No MCP? No problem. Every tool is a plain HTTP endpoint, and the catalog is self-documenting two ways: GET /tools returns names, descriptions, and JSON Schema inputs, while GET /openapi.json serves a full OpenAPI 3.1 spec — import it into anything that speaks OpenAPI (custom GPT actions, Open WebUI, LangChain toolkits, ...). Feed either to your assistant, POST its tool calls through, done.

# List all available tools (with their input schemas)
curl http://localhost:3000/tools -H "x-mcp-auth: YOUR_KEY"

# Get your courses
curl -X POST http://localhost:3000/tools/get-courses-summary \
  -H "x-mcp-auth: YOUR_KEY" -d '{}'

# Check your grades
curl -X POST http://localhost:3000/tools/get-grades-summary \
  -H "x-mcp-auth: YOUR_KEY" -d '{}'

# Read your inbox
curl -X POST http://localhost:3000/tools/get-mio-messages \
  -H "x-mcp-auth: YOUR_KEY" -d '{}'

The tools endpoint accepts JSON bodies regardless of Content-Type — no need to set it explicitly.

Straight to Omnivox (the proxy)

For anything the tools don't cover, /Mobl/* forwards requests directly to Omnivox's mobile API through the authenticated browser — any method, any content type, session cookies included:

curl -X POST http://localhost:3000/Mobl/LeaEtudiant/GetNotesDetailModel \
  -H "x-mcp-auth: YOUR_KEY" -H "Content-Type: application/json" \
  -d '{"AnSession":"20261","noCours":"2434K6EM","noGroupe":"1011"}'

Pair it with the endpoint notes in docs/ and you have the whole mobile API at your fingertips.

In your own projects

The gateway isn't just for AI — it's a full Omnivox API for anything that speaks HTTP:

• Grade dashboard — pull your grades into a web app or spreadsheet

• Schedule widget — your timetable on a home screen or desktop widget

• Assignment tracker — notifications for new assignments and deadlines

• Discord/Slack bot — get pinged when grades or messages land

• Mobile app — a custom Omnivox client with just the features you care about

• Data export — archive your grades, documents, and messages

Python, JavaScript, Swift, Kotlin, Go — whatever you like. Start from GET /tools.

The research corner

Omnivox's mobile API has no public documentation, so this repo doubles as a field guide. docs/ collects what we've mapped so far:

• Per-module endpoint docs — request bodies and response shapes for every /Mobl/ controller we've charted, from LÉA grades to MIO messaging.

• The service manifest — how GetOffreService describes the modules the app is assembled from, what the flags actually mean, and how its change-detection hash is built.

• Conventions — the ×100 fixed-point grade encoding, int-sentinel "no value" markers, term ID format, the common response envelope.

Some of it was found by reading the app's bundle, some by probing live endpoints — including a few natively-served endpoints the official app doesn't even use anymore (that's how per-evaluation grades work here without HTML scraping). If you chart new territory, contributions are warmly welcome.

Documentation

The Wiki is the operator's manual:

• Setup — prerequisites, installation, first login, Docker, VPS deployment, security precautions

• Connecting Assistants — MCP client configs, and non-MCP assistants via the tool catalog / OpenAPI spec

• Troubleshooting — common issues and fixes

For the API surface itself, the server is self-documenting: GET /tools and GET /openapi.json always reflect the running instance, and the Omnivox mobile API notes live in docs/.

Security

The short version: this behaves like a browser you left logged in on your own computer — nothing more, nothing less. The longer version, because it matters:

Your password and 2FA are untouched

• You log in yourself, on the real login page. The Electron app opens the official Omnivox login — password, 2-Step Verification, trusted-device prompt and all. This project never asks for, sees, stores, or replays your password or MFA codes.

• MFA keeps working. Omnivox's 2-Step Verification is fully compatible: you complete it once during login, and the resulting session (including the trusted-device cookie) is stored exactly as any browser would store it. Nothing here bypasses, weakens, or automates any Omnivox security mechanism — requests run through Omnivox's own JavaScript, so challenge-response auth and session expiry work exactly as Skytech designed them.

• Keep MFA enabled. This project works fine with it, and you should absolutely leave it on.

What's stored, and where

• Session cookies and browser profile — the data folder on your disk (data/ in the project, gitignored), same as a logged-in browser. Nothing is sent anywhere else.

• Access key — data/accessKey.txt, auto-generated, protects every endpoint the server exposes.

• No external servers, no telemetry. Everything runs on your machine or your server; all traffic goes directly between you and Omnivox. The codebase is open — audit it.

What the access key can do

Treat accessKey.txt like your Omnivox password. Anyone holding it gets your full session: reading grades and messages, sending MIO, downloading documents. If it ever leaks:

1. Delete accessKey.txt (a new key is generated on restart), and

2. npm run reset + re-login if you want to invalidate the session itself.

Hardening checklist

• Don't expose the server to the internet unless you actually need to. Localhost or LAN is the happy default. If you must go public: TLS via a reverse proxy (e.g. Nginx — sample config), and consider IP allowlisting.

• In Docker, bind to loopback unless you're fronting it with a proxy: "127.0.0.1:3000:3000" instead of "3000:3000".

• Keep the data folder private — it's gitignored already; add file permissions, and disk encryption if it's a laptop.

• One instance, one account, by design — there is nothing multi-tenant to leak across.

• More in Security Precautions.

A note for Skytech

This is a personal-use client for a student's own account — functionally a browser with an API, authenticating through your own login flow, MFA included, one account per instance. It doesn't harvest credentials, doesn't circumvent any protection, and can't see anything the logged-in student can't already see in the portal themselves. If you're from Skytech or a college and have concerns, please open an issue — happy to talk.

Disclaimer

This project is unofficial and not affiliated with, endorsed by, or associated with Skytech Communications, Omnivox, or any Quebec college. It is provided as-is for personal and educational use only. Use it at your own risk — the authors are not responsible for any consequences resulting from its use, including but not limited to account restrictions, data loss, or violations of your institution's terms of service.

License

ISC