schoolbridge
Connects Canvas LMS (and other school platforms) to AI agents, enabling them to monitor assignments, grades, announcements, and changes, then rank work, brief students, or build study plans via MCP tools.
README
schoolbridge
Connect Canvas (and other school platforms) to AI agents.
schoolbridge watches your school's LMS and turns it into something an AI agent can actually use. When a teacher posts an assignment, uploads a grade, changes a due date, or makes an announcement, your agent finds out — and can brief you, rank your week, or build a study plan for the next test.
It speaks three dialects, so it works with almost any agent setup:
| Interface | For | Example |
|---|---|---|
| MCP server | Claude Code, Claude Desktop, Claude Cowork, any MCP client | schoolbridge mcp |
Agent skill (SKILL.md) |
Hermes, OpenClaw, any Agent-Skills runtime | schoolbridge install-skill hermes |
| JSON CLI | Shell-driven agents, cron jobs, scripts | schoolbridge upcoming --json |
| Event watcher | Push-style automations | schoolbridge watch --exec 'my-agent brief' |
Canvas is the built-in provider. The provider interface is small and documented, so Google Classroom, Schoology, Moodle, and others can be added — see docs/PROVIDERS.md.
schoolbridge is read-only: it never writes anything back to your LMS.
Quick start
npm install -g schoolbridge # or run everything below with: npx schoolbridge …
One-line install (ideal for VPSes and agent hosts — installs the CLI, connects Canvas, and installs the agent skill in one shot):
curl -fsSL https://raw.githubusercontent.com/Shoberman2/schoolbridge/main/install.sh | bash -s -- \
--base-url https://yourschool.instructure.com --token "<canvas-token>" --skill hermes
(All flags optional — bare | bash just installs the CLI. --skill accepts hermes, openclaw, or agents.)
Try it instantly (no credentials)
Every command accepts --provider mock, which serves realistic sample data:
schoolbridge upcoming --provider mock
1. [CRITICAL 73] Chapter 12 Reading Quiz
US History · overdue by 20h · 10 pts · missing
2. [HIGH 66] Unit 4 Test: Cellular Energetics
AP Biology · due in 4d 23h · 100 pts · unsubmitted
3. [HIGH 59] Reconstruction DBQ Essay
US History · due in 2d 23h · 100 pts · unsubmitted
...
Connect your real Canvas account
- Log in to Canvas in a browser → Account → Settings → scroll to Approved Integrations → + New Access Token. Copy the token immediately.
- Run:
schoolbridge init --base-url https://yourschool.instructure.com --token <paste-token>
init verifies the connection, then saves the config to ~/.schoolbridge/config.json (created with 0600 permissions — it holds your token). You can also skip the file entirely and use the CANVAS_BASE_URL and CANVAS_ACCESS_TOKEN environment variables.
Can't get a token? Use the zero-credential calendar feed
Some schools disable student access tokens (and the Canvas mobile app doesn't have the token button at all). Every Canvas account still has a calendar feed that needs no token and no admin:
- Open Canvas in a web browser → Calendar
- Click "Calendar Feed" (bottom-right of the page) and copy the URL (ends in
.ics) - Run:
schoolbridge init --provider ics --feed-url <paste-url>
The ics provider covers assignments, due dates, and calendar events — so upcoming (with priority ranking), calendar, study planning, and change events (new_assignment, due_date_changed, new_calendar_event, calendar_event_changed) all work. The feed carries no grades, announcements, submissions, or feedback, so those surfaces stay empty until you can add a token or OAuth.
Third-party apps: requesting Canvas API permission (OAuth2)
Personal tokens are fine for your own agent on your own machine. When schoolbridge is embedded in a third-party app or service, the user shouldn't hand over a raw token — Canvas's answer is OAuth2, where the app requests permission and the user approves it on Canvas's own consent screen. schoolbridge implements the full flow:
-
Get a Developer Key. The app developer asks the school's Canvas admin to create one (Admin → Developer Keys → + API Key) with redirect URI
http://localhost:8765/oauth/callback. The admin controls whether to approve, and can scope/revoke the key at any time. This yields a client id and client secret. -
The user authorizes in their browser:
schoolbridge auth login --base-url https://yourschool.instructure.com \ --client-id 10000000000001 --client-secret <secret>schoolbridge opens Canvas's consent page, catches the callback on localhost (with CSRF
statevalidation), exchanges the code for tokens, verifies the connection, and stores the session. -
Tokens refresh automatically. Access tokens expire hourly; schoolbridge refreshes them transparently (and retries once on a 401), so agents and watchers keep running indefinitely.
Manage the session with schoolbridge auth status and schoolbridge auth logout (which revokes the token with Canvas, not just locally). Users can also revoke access anytime from Canvas → Account → Settings → Approved Integrations. A manual token (flag, env var, or init) always takes precedence over a stored OAuth session, and everything stays read-only either way.
Then:
schoolbridge upcoming # ranked work due this week (+ recent overdue work)
schoolbridge grades # course grades + recently graded assignments
schoolbridge announcements # recent teacher announcements
schoolbridge calendar # upcoming course calendar events
schoolbridge feedback # recent teacher comments on your work
schoolbridge events # everything that changed since the last check
Use with Claude (MCP)
Claude Code
claude mcp add schoolbridge -- npx -y schoolbridge mcp
Claude Desktop / Claude Cowork
Add to your MCP configuration (e.g. claude_desktop_config.json):
{
"mcpServers": {
"schoolbridge": {
"command": "npx",
"args": ["-y", "schoolbridge", "mcp"],
"env": {
"CANVAS_BASE_URL": "https://yourschool.instructure.com",
"CANVAS_ACCESS_TOKEN": "your-token"
}
}
}
}
(If you ran schoolbridge init, the env block is optional — the server reads the saved config.)
Then just ask:
"What's new at school?" · "Rank everything I have due this week and plan my days." · "Make me a study plan for the biology test."
MCP tools
| Tool | Returns |
|---|---|
list_courses |
Active courses with current grade/score |
list_upcoming_work |
Work due in the next N days (default 7) + recent overdue work, each with a 0–100 priority ranking hint |
get_assignment_details |
One assignment with its full instructions as plain text |
list_announcements |
Teacher announcements from the last N days |
list_calendar_events |
Course calendar events for the next N days |
list_recent_feedback |
Teacher comments on your submitted work |
get_grades |
Course grades + everything graded in the last two weeks |
check_new_events |
Everything that changed since the last check (see Events) |
MCP prompts
| Prompt | Does |
|---|---|
whats_new |
Check for changes and brief the student |
plan_my_week |
Rank the week's work and lay out a day-by-day plan |
study_plan |
Build a day-by-day study plan for a test (args: course, optional assignment) |
Use with Hermes
schoolbridge ships a portable Agent Skills SKILL.md that teaches Hermes the whole workflow — setup, ranking, study plans, and event monitoring. Install it either way:
npm install -g schoolbridge
schoolbridge install-skill hermes # writes ~/.hermes/skills/schoolbridge/SKILL.md
or straight from this repo with Hermes' own installer:
hermes skills install https://raw.githubusercontent.com/Shoberman2/schoolbridge/main/skill/SKILL.md
Hermes auto-discovers the skill on startup and activates it whenever the conversation is school-shaped ("what homework do I have?", "plan my week", "make a study plan for the bio test"). Add schoolbridge events --json to a heartbeat/schedule and Hermes will proactively tell you when a teacher posts an assignment, uploads a grade, or makes an announcement.
Use with OpenClaw
Same skill, OpenClaw flavor — the frontmatter carries an metadata.openclaw block that gates on the schoolbridge binary and points OpenClaw's installer at the npm package:
npm install -g schoolbridge
schoolbridge install-skill openclaw # writes ~/.openclaw/skills/schoolbridge/SKILL.md
Per-workspace instead: copy skill/ into <workspace>/skills/schoolbridge/. For proactive alerts, add a line to your HEARTBEAT.md such as: "Run schoolbridge events --json; if it prints events, tell me about them."
There's also schoolbridge install-skill agents for the shared ~/.agents/skills directory used by other Agent-Skills-compatible runtimes.
Use with any other agent (shell/JSON)
Every read command takes --json and prints clean, stable JSON — pipe it straight into your agent's context:
schoolbridge upcoming --json --days 7
schoolbridge grades --json
schoolbridge announcements --json --days 3
schoolbridge assignment 101 5002 --json # full details incl. instructions
Polling for changes
schoolbridge events diffs the LMS against the previous run and prints only what changed. The first run saves a baseline and prints nothing. State lives in ~/.schoolbridge/state.<provider>.json.
schoolbridge events --json # one JSON event per line; empty output = nothing new
schoolbridge events --reset # start over from a fresh baseline
A cron heartbeat for a shell-based agent:
*/15 8-22 * * * schoolbridge events --json | your-agent ingest-school-events
Push mode
watch polls on an interval, prints events to stdout as JSON lines (logs go to stderr), and can push each batch onward:
schoolbridge watch --interval 15m --exec 'your-agent brief --stdin' # JSON payload on stdin
schoolbridge watch --interval 15m --webhook https://your-agent.example/hooks/school
The webhook/exec payload:
{
"source": "schoolbridge",
"provider": "canvas",
"generatedAt": "2026-08-17T20:15:00.000Z",
"events": [ … ]
}
Events
Every event has the same shape, so agents can pattern-match on type:
{
"type": "grade_posted",
"occurredAt": "2026-08-17T20:15:00.000Z",
"courseId": "101",
"courseName": "AP Biology",
"title": "Cell Respiration Lab Report",
"summary": "Grade posted in AP Biology: “Cell Respiration Lab Report” — 47/50 (94%).",
"url": "https://yourschool.instructure.com/courses/101/assignments/5001",
"data": { "assignmentId": "5001", "score": 47, "grade": "47", "pointsPossible": 50 }
}
type |
Fires when |
|---|---|
new_assignment |
A teacher posts an assignment, quiz, or test |
due_date_changed |
An assignment is rescheduled |
grade_posted |
A grade appears on previously ungraded work |
grade_changed |
An existing grade is revised |
new_announcement |
A teacher posts an announcement |
new_discussion |
A teacher opens a discussion topic |
new_calendar_event |
Something lands on a course calendar (review session, field trip, in-class test…) |
calendar_event_changed |
A calendar event is rescheduled |
new_module_item |
New course content is published (pages, linked resources…) |
new_file |
A file is added to a course |
new_feedback |
A teacher comments on your submitted work |
course_grade_changed |
Your overall course grade moves |
Coverage degrades gracefully: if your institution disables a surface (e.g. the Files tab), that category is silently skipped rather than erroring. When upgrading schoolbridge, newly added categories baseline quietly on the first poll instead of flooding you with "new" events for things that already existed.
summary is always a ready-to-speak sentence; data carries the structured before/after values.
Priority ranking
list_upcoming_work / schoolbridge upcoming attach a priority score (0–100) and label (critical / high / medium / low) to each item. It weighs:
- due-date proximity (heaviest factor),
- point value,
- test-likeness (quizzes, or names matching test/exam/midterm/final),
- missing/overdue status (boost) and already submitted (drops to ~0).
It's deliberately a hint, not a verdict — the intended pattern is for the AI to use it as a starting order and override it with judgment (start the essay before the worksheet, even if the worksheet is due first).
Use as a library
import { CanvasProvider, listUpcoming, checkEvents, StateStore } from "schoolbridge";
const provider = new CanvasProvider({ baseUrl: "https://yourschool.instructure.com", token: process.env.CANVAS_ACCESS_TOKEN! });
const ranked = await listUpcoming(provider, 7);
const { events } = await checkEvents(provider, new StateStore(provider.name));
Configuration reference
Resolution order for every setting: CLI flag → environment variable → ~/.schoolbridge/config.json.
| Setting | Flag | Env var |
|---|---|---|
Provider (canvas / mock) |
--provider |
SCHOOLBRIDGE_PROVIDER |
| Canvas URL | --base-url |
CANVAS_BASE_URL |
| Canvas token | --token |
CANVAS_ACCESS_TOKEN (or CANVAS_TOKEN) |
| Config/state directory | — | SCHOOLBRIDGE_HOME (default ~/.schoolbridge) |
Privacy & safety
- Your token and all state stay on your machine; schoolbridge talks only to your school's Canvas host (and, if you opt in, your own webhook).
- All operations are read-only against the LMS.
- Treat your Canvas token like a password. You can revoke it any time from Canvas → Account → Settings.
Roadmap
- Providers: Google Classroom, Schoology, Moodle, PowerSchool
- Calendar export (ICS) of upcoming work
- Native push (Canvas live events) where institutions allow it
Contributing
PRs welcome — especially new providers. See CONTRIBUTING.md and docs/PROVIDERS.md. The mock provider and npm test let you develop without any school credentials.
License
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.