For the complete documentation index, see llms.txt. This page is also available as Markdown.

Troubleshooting

Fix common Navixy MCP Server problems across AI clients, including disconnected servers, missing tools, 401 errors, and account switching.

This page covers known issues and solutions for Navixy MCP Server across all supported AI clients.

How do I log out or switch accounts?

  • Claude Desktop: Delete the .mcp-auth\mcp-remote-xxxx folder in your user directory. This logs out all MCP servers simultaneously.

  • Cursor: Go to ⚙️ Cursor Settings → Tools and click Logout next to the Navixy server.

  • ChatGPT web: Go to Settings → Plugins, select the Navixy plugin, click , and select Reconnect.

  • Codex: Run codex mcp logout navixy-user, then run codex mcp login navixy-user.

"MCP {name}: Server disconnected" in Claude Desktop

This error appears in Claude Desktop when the MCP server connection is lost.

To resolve this:

  1. Restart Claude Desktop completely via ≡ → Files → Exit. A window reload is not enough.

  2. If restarting doesn't help, delete the .mcp-auth\mcp-remote-xxxx folder in your user directory and reconnect.

  3. If that doesn't help either, reboot your operating system.

  4. If that still fails, it might be a Node.js/npm issue. Try installing pnpm and replacing npx in claude_desktop_config.json with pnmp.

Claude Desktop can't find npx on macOS

If you encounter issues with Node.js on macOS, install it system-wide via Homebrew:

"I don't see any connected Navixy resources or tools available" in ChatGPT

This error appears in ChatGPT even when the MCP server has not been disconnected. The most common cause is that Developer mode is turned off.

To resolve this:

  1. Open Settings → Security and login.

  2. Confirm that Developer mode is enabled. If it is off, turn it on.

  3. Start a new conversation.

  4. Select the Navixy plugin from the Developer mode tool in the composer.

A 401 error means that the MCP server on your computer doesn't have valid sign-in information. Try to authenticate in the ChatGPT desktop app first.

  1. Open Settings → MCP servers. In some versions, open Settings → Plugins → MCPs.

  2. Find navixy-user or navixy-panel in the MCP server list.

  3. Click Authenticate and enter your Navixy credentials.

  4. Restart the ChatGPT desktop app.

  5. Open a Codex conversation and enter /mcp.

The server is ready when /mcp shows it as connected.

Use the command-line fallback

Use this fallback only if Authenticate doesn't appear or the 401 error continues.

  1. Open Terminal on macOS or PowerShell on Windows.

  2. Run codex --version to check whether Codex CLI is available.

  3. Run the login command for your connection.

For Navixy User MCP, run the following command:

For Navixy Panel MCP, run the following command:

Enter your Navixy credentials in the browser, then restart the ChatGPT desktop app. Enter /mcp in a Codex conversation to confirm the connection.

If the terminal reports that the server doesn't exist, add and authenticate it with the matching commands:

If the codex command isn't available

On macOS, the ChatGPT desktop app includes Codex CLI inside the app. Use the following path in place of codex:

Some upgraded installations use /Applications/Codex.app/Contents/Resources/codex instead.

On macOS or Windows, you can also install Node.js and then install Codex CLI:

See Configuring Navixy MCP in ChatGPT and Codex for the complete setup.

Codex shows an unsupported model error

The ChatGPT desktop app and Codex CLI share the model setting in ~/.codex/config.toml. An unavailable model can prevent a Codex conversation from starting.

Use the model control below the desktop composer to select an available model. In Codex CLI, enter /model and select an available model.

If the error continues, open ~/.codex/config.toml. Remove or update the model value. For example, gpt-5.2-codex may not be available with ChatGPT authentication.

The assistant answers but does not cite sources

Ask for links explicitly. You can also ask it to quote a short phrase from the section it referenced to confirm it retrieved the right content.

Pasted content looks broken

Check for wrapped lines. This is most common with curl commands and long API paths copied from the docs.

Last updated

Was this helpful?