folava MCP

Guide / Use in ChatGPT

On this page

Use Folava in ChatGPT

Connect Folava's tools to a ChatGPT web conversation through a private MCP tunnel. Folava runs on your computer; the project website is documentation, not a server URL to enter in ChatGPT.

This recipe follows current OpenAI documentation. A complete Folava connection and tool call in ChatGPT have not yet been verified.

1. Start with the demo

Install the published release. The first connection uses folava-mcp serve --demo: fictional coursework, with no Canvas credentials or Canvas requests.

You need permission to add custom MCP plugins in ChatGPT and to create/use a Platform tunnel. Availability depends on your account and workspace.

2. Connect a private tunnel

In Platform tunnel settings, use your personal Platform organization and associate the tunnel with your personal ChatGPT workspace. Creation needs Tunnels Read + Manage; runtime access needs Read + Use. Keep access limited to yourself before connecting Canvas; a shared workspace association can give others a path to this process.

Install OpenAI's tunnel-client using its official tunnel guide. Supply a restricted runtime key as CONTROL_PLANE_API_KEY through your local secret manager. This is an OpenAI key, separate from Canvas credentials. Keep literal keys outside chat, shell history and configuration files.

Replace YOUR_TUNNEL_ID with the ID from settings. Use an unused profile name:

tunnel-client init \
  --sample sample_mcp_stdio_local \
  --profile folava-demo \
  --tunnel-id YOUR_TUNNEL_ID \
  --mcp-command "folava-mcp serve --demo"
tunnel-client doctor --profile folava-demo --explain
tunnel-client run --profile folava-demo

If needed, use Folava's full executable path from uv tool dir --bin, including .exe on Windows. Run both processes as the user who installed Folava.

Leave the client running. This route uses an outbound connection without a public Folava listener. See the official guide for platform-specific setup.

3. Add it to ChatGPT

In ChatGPT Plugins:

  1. Select +, then Add custom MCP server.
  2. Name it Folava demo, choose Tunnel under Connection and select your tunnel.
  3. For this local stdio connection, choose No authentication. Access is controlled by the private tunnel; this is separate from Canvas sign-in.
  4. Review the warning, create it under your personal plugins and install it.

In a new conversation, type @, select Folava demo, and ask:

Use Folava demo to list my courses with canvas_courses.

Expect fictional course ENV101. Confirm an actual tool response, rather than just a successful plugin installation. The OpenAI connection instructions describe the current controls.

4. Connect your Canvas account

Read the Canvas authentication guide first, including the approval and institution requirements. Authenticate privately and run:

folava-mcp doctor --profile default

Stop the demo tunnel client. Create a separate tunnel-client profile using the same tunnel ID, changing the profile name to folava-canvas and the MCP command to folava-mcp serve --profile default. Run its doctor check, then start it. Rename the personal ChatGPT plugin to Folava and refresh its tool list.

Folava enforces the same reviewed read boundary as in other MCP hosts. Keep both tunnel and plugin access limited to your own account: callers use the Canvas profile available to that process. Course results sent into ChatGPT are subject to OpenAI's data terms; see data handling.

If it does not connect

  • Tunnel missing: check its workspace association and your Use permission.
  • Tool discovery fails: keep the host awake and the client running; check tunnel-client doctor --profile folava-demo --explain (or folava-canvas).
  • Demo works, Canvas fails: run Folava's doctor as the same OS user. Check that the real profile, executable path and credential store are available.

Stopping the tunnel client stops this connection. Remove the plugin and tunnel when no longer needed; these steps do not revoke a Canvas credential.

Using the desktop app

The official MCP settings guide also describes local servers in the ChatGPT desktop app: Settings → MCP servers → Add server, choose STDIO, command folava-mcp, arguments serve --demo, then save and restart. Type /mcp to check the connection. After authentication, use serve --profile default for Canvas.