Getting started

Let your agent help

Send your agent this setup link and ask it to install Folava MCP. It can install the published release, preserve your existing MCP configuration and connect the synthetic demo. Your account connection happens separately, with credentials kept outside the conversation. The homepage has a ready-to-copy prompt; the manual steps are below.

For a ChatGPT conversation, follow Use Folava in ChatGPT. The web guide uses a private tunnel to this local process; it does not require publishing your Canvas connection as a public server.

You need Python 3.11 or newer and an MCP host that can launch a stdio command. uv manages an isolated tool installation on Linux, macOS and Windows. No Node runtime is required.

Install

git clone https://github.com/liltaket/folava-mcp.git
cd folava-mcp
uv tool install .
folava-mcp --help

For an immutable installation, check out the release tag before installing, or install the wheel attached to the GitHub release. Package versions are pinned in uv.lock for development and CI. The application is not claimed to be on PyPI. To update a source installation, update the checkout and use uv tool install . --force. Review changes before upgrading credentials-handling software.

Codex

Try the synthetic workspace first:

codex mcp add canvas-demo -- folava-mcp serve --demo

Ask for the demo course's assignments or field guide. Demo dates are fixed sample data, not relative to today. The process waits for MCP messages when run directly.

After authentication and folava-mcp doctor:

codex mcp add canvas -- folava-mcp serve --profile default
codex mcp list

If your GUI client cannot find the command, use the absolute executable path reported by uv tool dir --bin. On Windows include .exe where required.

The equivalent configuration is available with folava-mcp config:

[mcp_servers.folava]
command = "folava-mcp"
args = ["serve", "--profile", "default"]

Codex supports local stdio servers and stores this in its MCP configuration; see official Codex MCP documentation. Review the generated snippet and merge it into your existing configuration; the CLI does not overwrite your file. A codex mcp login command authenticates an HTTP MCP server, not this local server's upstream Canvas connection.

T3 Code and multiple computers

Configure the Codex provider on the machine where its Codex process executes. When T3 runs the provider on another linked computer, that computer needs the executable, credentials and Codex MCP entry. Restart the provider/session after configuration changes. There is no assumed T3-specific Canvas plugin or automatic configuration propagation in this release. External T3 UI behavior has not been validated here; the official SDK protocol test is a separate check.

Repeat installation and institution-approved authentication per computer. Use separate named profiles for separate accounts. Avoid copying tokens into shared project configuration. Centralized authenticated HTTP hosting could reduce per-device setup, but is intentionally outside this release's supported surface.

Other MCP hosts

folava-mcp config --format json prints:

{
  "mcpServers": {
    "folava": {
      "command": "folava-mcp",
      "args": ["serve", "--profile", "default"]
    }
  }
}

Merge that entry into a host's supported stdio configuration. Host-specific paths and approval interfaces vary. The server exposes tools as well as a guide resource so ordinary workflows do not depend on a resource browser.