Skills
The three Canvas agent skills, how they differ from the MCP server, and how to install them.
Server vs. skill
The MCP server is the hands: it holds your token and makes the actual HTTP calls to Canvas. A skill is the playbook: a written, multi-step procedure that tells the agent which tools to call, in what order, and how to reason about the results — without adding any new access to Canvas itself.
You need the server installed either way — see Install. Skills
don't replace it or talk to Canvas on their own; they sit on top of the 17 tools
already described in the tool reference and give the agent a
named, repeatable workflow to run against them. Without a skill, an agent can still
chain whats_due → my_courses → my_submission on its own if asked the right way;
a skill just writes that chain down once so it happens the same way every time,
triggered by a shorter, more natural request.
Install
npx skills add JohannsenLum/canvas-api-mcpThis reads the skill definitions straight from the GitHub repository — there's no npm package to publish and no separate registry involved. Because it's just files read off GitHub, the language the server itself is written in (Python) is irrelevant to this step.
The three skills
canvas-week-plan
Merges upcoming deadlines with your current submission state and flags what's
urgent — due soon and unsubmitted, or several items landing the same day — walking
through whats_due, my_courses, and my_submission in sequence rather than just
listing raw due dates.
"What should I prioritise this week on Canvas?"
canvas-study-pack
Gathers modules, pages, and files for a given course and topic into a single study
set, walking through my_courses, course_content, list_files, and read_file.
The summary it produces is grounded only in material it actually found — it's told
not to invent coverage.
"Put together a study pack for CS3230 on dynamic programming."
canvas-grade-check
Computes your current standing in a course and what the remaining ungraded work is
worth toward it, walking through my_courses, my_grades, and list_assignments.
"How am I doing in CS2103T, and what's left that could still move my grade?"
Prompts do the same job for MCP-native clients
These three workflows also ship as MCP prompts — week_ahead, study_pack, and
grade_check, listed in the tool reference — built into
the server itself rather than installed separately. Any MCP client that surfaces
prompts already has this behaviour with no extra install step; the skills exist for
clients that support Agent Skills but don't surface MCP prompts. Use whichever route
your client gives you — they walk the same tool sequence either way.