CoursePack Local MCP Server
Read-only MCP server that provides Claude Desktop with access to converted Canvas course exports, including Markdown files, reports, and search tools for local course content.
README
CoursePack Local MVP v9
CoursePack Local converts Canvas .imscc course exports into a clean, local Markdown coursepack. It now includes a local web interface, course viewer, search, safe skip logging, and a read-only MCP server connection for Claude Desktop.
The project is still an MVP, but v9 is closer to a deployable instructor workflow:
- Double-click
Start CoursePack Local.bat. - Browser opens at
http://127.0.0.1:3333. - Upload a Canvas
.imsccexport. - Review the converted course, skipped files, and report.
- Search the converted course.
- Connect the local read-only MCP server to Claude Desktop.
What changed in v9
- Added a stronger Recent Courses page with last-updated information, quick actions, and local delete buttons.
- Added built-in local reports that do not require AI: date findings, link findings, missing local links, policy mentions, skipped-file summary, and workload by module.
- Reports are saved as
reports/reports.mdandreports/reports.jsonin each converted course. - Added a Help page explaining how instructors open CoursePack again after the first install.
- Added a user-level uninstall helper:
Uninstall CoursePack Local.batanduninstall.ps1. - Added MCP tools for Claude to read the built-in reports.
- Kept v8 behavior: the one-line installer does not auto-connect Claude Desktop during installation.
- Kept v8 behavior: installer starts CoursePack first, opens the local browser page when possible, and creates Desktop/Start Menu shortcuts.
- Kept v8 behavior: Claude connection has immediate browser feedback and is idempotent.
What it does now
- Reads a Canvas
.imsccfile as a ZIP archive. - Parses
imsmanifest.xmlandcourse_settings/module_meta.xmlwhen present. - Converts Canvas wiki HTML pages to Markdown.
- Converts assignment HTML pages and assignment metadata.
- Converts discussion/announcement XML topic bodies.
- Extracts basic quiz/QTI text.
- Builds module index files.
- Converts loose instructor-uploaded text-like files, such as
.html,.txt,.md,.csv, and.json. - Skips images, audio, video, PowerPoint, Office files, PDFs, unknown files, and very large files by default.
- Can optionally attempt PDF/Office conversion with Microsoft MarkItDown if the optional dependency is installed.
- Writes a human-readable
conversion_report.md. - Writes structured metadata in
metadata/course_map.json,metadata/skipped_assets.json,metadata/conversion_events.json, andmetadata/resource_map.json. - Runs a local web app for upload, review, search, reports, and download.
- Provides built-in rule-based reports that work without AI.
- Provides a read-only MCP server for Claude Desktop.
Windows quick start
- Unzip the CoursePack folder.
- Double-click:
Start CoursePack Local.bat
The first run creates a local .venv and installs required packages.
- Open or wait for the browser page:
http://127.0.0.1:3333
- Upload a Canvas
.imsccexport and click Convert Course.
Using CoursePack after the first install
Instructors do not run the one-line installer every time. The one-line command is only for first install or reinstall/update.
After CoursePack is installed, use one of these:
Desktop shortcut: CoursePack Local
Start Menu shortcut: CoursePack Local
Manual address after starting: http://127.0.0.1:3333
Converted courses and app data are stored in the user's local profile:
%LOCALAPPDATA%\CoursePackLocal
If CoursePack is already running and the instructor clicks the shortcut again, the app should simply open the browser page instead of failing because the port is already in use.
Claude connection timing
CoursePack does not need a converted course before the MCP connector can be registered with Claude Desktop, but Claude will not have useful course content until at least one Canvas export is converted.
Recommended flow:
1. Install CoursePack.
2. Open the local CoursePack page.
3. Convert a Canvas export.
4. Click Claude Desktop > Connect CoursePack to Claude Desktop.
5. Fully quit and reopen Claude Desktop.
The installer intentionally does not auto-connect Claude anymore. This prevents a Claude/Windows security problem from blocking the app startup.
Check the install
After running CoursePack once, double-click:
Check CoursePack Install.bat
This checks Python, required packages, local app files, and Claude configuration readability.
To remove the local app later, use the Start Menu shortcut or run:
Uninstall CoursePack Local.bat
Connect to Claude Desktop
Windows note: Standalone Claude and Microsoft Store Claude use different claude_desktop_config.json files. CoursePack's Connect button updates all relevant config files so the Claude you actually launch can see the coursepack MCP tools.
The easiest method is from the browser UI:
- Open CoursePack Local.
- Go to Claude Desktop in the top navigation.
- Click Connect CoursePack to Claude Desktop.
- Fully quit Claude from the system tray (Quit — not just close the window).
- Reopen Claude and look for
coursepackMCP tools.
If tools are still missing on Microsoft Store Claude, check:
%LOCALAPPDATA%\Packages\Claude_*\LocalCache\Roaming\Claude\logs\mcp.log
Claude Desktop reads local MCP servers from its config file(s). CoursePack adds a coursepack MCP server entry pointing to the packaged .exe (--mcp) or local Python + mcp_server.py in source mode.
What Claude can access
The MCP server is read-only. It exposes:
- converted Markdown course files
- course map metadata
- conversion report
- skipped assets log
- conversion events log
- local search tools
It does not write back to Canvas.
Failure behavior
A single bad file should never stop the full course conversion.
The converter records:
- converted files in
metadata/course_map.json - skipped files in
metadata/skipped_assets.json - warnings/errors in
metadata/conversion_events.json - a readable summary in
conversion_report.md
If a document cannot be converted, CoursePack skips it, logs it, and keeps converting the rest of the course.
Optional document conversion
By default, PDFs and Office files are skipped because they can be large, slow, or messy.
To attempt conversion of PDF/Office files, install the optional converter dependency:
pip install -r requirements-docs.txt
Then enable the optional document conversion checkbox in the upload page.
If MarkItDown is not installed, the converter will skip those files, record the reason, and continue.
Media files skipped by default
Skipped by default:
- images:
.jpg,.jpeg,.png,.gif,.webp,.svg,.bmp,.tif,.tiff,.heic - audio:
.mp3,.wav,.m4a,.aac,.ogg,.flac - video:
.mp4,.mov,.avi,.mkv,.webm,.wmv,.m4v - PDF/Office files unless optional conversion is enabled
- unknown file types
- large files above the conversion size limit
Current deployment limitations
This is closer to deployment, but not fully packaged for nontechnical instructors yet.
Source/dev mode still requires:
- Python installed on the computer
- first-run package installation into
.venv - Windows batch files for startup
Portable/release mode bundles Python with PyInstaller. Some managed university computers may still block unknown unsigned executables. In that case, use the source/dev mode or ask IT to allow/sign the executable.
Recommended next steps
- Test v9 on Windows with and without Claude Desktop installed.
- Test with several Canvas exports from different course types.
- If managed computers block the unsigned portable executable, add code signing or an IT-deployed MSI.
- Add optional AI chat inside the local browser UI.
- Add a Claude Desktop
.mcpbextension package later for the cleanest Claude install experience.
One-line Windows installer
Recommended (window stays open on managed PCs):
powershell -NoProfile -ExecutionPolicy Bypass -NoExit -Command "iex (irm 'https://raw.githubusercontent.com/digaocoite/canvasmcplocal/main/install.ps1')"
Repo: digaocoite/canvasmcplocal. See ONE_LINE_INSTALL.md for details.
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.