Fixing a stuck connection.
Most connection problems clear up with a reconnect: the same steps as the first time, from the Setup guides page.
Stuck in a browser-approve loop
The client opens the consent screen, you approve, and it opens again immediately. This usually means the client didn't store the returned grant. Check that your system clock is correct (OAuth tokens are time-bound) and that the client has write access to its own config/state directory (some corporate device-management tools sandbox this).
Tool calls suddenly return 401, but it worked before
Expected after about 30 days of the connection sitting idle: the refresh token expired. Reconnect using the same steps on the Setup guides page; it re-opens the browser approval and issues a fresh grant. If it happens sooner than 30 days, the connection may have been revoked from Account → API keys. Reconnect the same way.
The consent screen never opens
Some remote/headless environments (SSH sessions, containers, CI runners) have no browser to open. MCP OAuth needs a real browser on the same machine as the client. Run the client locally to connect the first time, or use a manual API key (Account → API keys) for headless/CI use instead.
Behind a corporate proxy or firewall
Both the initial tool call and the OAuth redirect need outbound HTTPS to Ordel's domain. If your network requires an explicit proxy, configure it at the client level (most clients honor HTTPS_PROXY / HTTP_PROXY) rather than in Ordel's MCP URL. The URL itself never changes.
Reconnect from each client's own MCP menu
Every client keeps its own place to see and reset an MCP connection. Use it instead of editing config files by hand once Ordel is already added:
Claude Code
If Claude Code stops calling Ordel tools, run `/mcp` and reconnect `ordel`. This re-opens the browser approval. Being asked to re-approve after about 30 days of inactivity is expected, not an error.
Cursor
If the connection drops, open Settings → MCP and reconnect `ordel`. This re-opens the browser approval.
Codex
Run `codex mcp login ordel` to force re-authentication. This is Codex's equivalent of reconnecting from a client MCP menu.
GitHub Copilot
Accounts menu → "Manage Trusted MCP Servers" (or "MCP: List Servers" → Restart) to re-authenticate.
Antigravity
Manage MCP Servers → "View raw config", or delete the stored entry in ~/.gemini/antigravity/mcp_oauth_tokens.json and reconnect to force re-approval.
Still stuck? Contact us with the client, its version, and what you see in its MCP log/output panel.