Skip to main content
Canvas MCP is a FastMCP server that makes your Canvas LMS data accessible to any AI assistant that speaks the Model Context Protocol. It exposes all 26 Canvas tools as MCP-callable functions, so clients like Claude Desktop and Cursor can list your courses, read assignments, fetch discussion threads, and more — without you writing a single line of glue code.

​
The canvas-mcp entry point

When you install the project, pyproject.toml registers a canvas-mcp console script:
pyproject.toml
[project.scripts]
canvas-mcp = "mcp_entry:main"
Running uv run canvas-mcp starts the FastMCP server. By default it uses the stdio transport, which is what desktop MCP clients expect. You can switch transports with the --transport flag — see Transports for the full list.

​
How tools are registered

The server uses dynamic registration rather than hand-written tool functions. At startup, mcp/server.py iterates over every entry in TOOL_SPECS — a registry of ToolSpec definitions that each carry a name, description, and JSON Schema parameter spec — and builds a real Python function for each one on the fly:
mcp/server.py
for tool_spec in TOOL_SPECS:
    _register_tool(tool_spec)
Inside _register_tool, the server:
  1. Derives Python type annotations from the JSON Schema (string → str, integer → int, boolean → bool, array → list[...]).
  2. Constructs a proper inspect.Signature so FastMCP can generate accurate MCP tool schemas.
  3. Attaches the spec’s description as __doc__ and registers the function via mcp.tool(...).
Optional parameters are typed as T | None with a None default; required parameters carry no default and appear as mandatory in the MCP schema.

​
Available tools

All 26 tools are registered automatically from TOOL_SPECS:
ToolDescription
get_todayToday’s date in ISO format
list_coursesList courses, with optional search and favorites filter
resolve_courseFuzzy-match a course by name
get_course_overviewCourse-level metadata and teacher info
get_course_syllabusSyllabus metadata and optional body
list_course_assignmentsList assignments with bucket filtering
get_assignment_detailsFull assignment details
list_course_pagesList course wiki pages
list_course_tabsList left-sidebar navigation tabs
get_course_tabFetch one tab and resolve its destination
canvas_get_pageGet a specific wiki page by URL or ID
list_discussion_topicsList discussions with title/search filters
get_discussion_entriesFetch thread entries and replies
list_course_filesList files in a course
download_course_fileDownload a file to local temp storage
list_course_foldersList folders
list_assignment_groupsList grading groups and optional assignments
list_course_submissionsList submissions for a student
get_course_grade_summaryGrade summary with assignment-group breakdown
list_modulesList modules with optional items
list_announcementsList announcements across courses
list_calendar_eventsList calendar events
list_course_peopleList users enrolled in a course
resolve_canvas_urlParse a Canvas URL and optionally fetch details
get_course_context_snapshotAggregated course context for quick orientation

​
What MCP clients can do

Any MCP-compatible AI assistant can connect to Canvas MCP and call these tools in natural language. Common workflows include:
  • Asking an assistant to summarize upcoming assignments across all your courses
  • Having Claude pull a discussion thread and draft a reply
  • Using Cursor to fetch a course’s grade summary while you work on a related project

Transports

Learn about stdio, HTTP, and SSE transports and when to use each one.

Client setup

Connect Canvas MCP to Claude Desktop, Cursor, or any HTTP-based client.

Build docs developers (and LLMs) love

Get started for freeTalk to us