MCP troubleshooting

Public endpoints

• MCP: https://api.agentmailer.ai/mcp
• OAuth protected-resource metadata: https://api.agentmailer.ai/.well-known/oauth-protected-resource
• Canonical signup and connection guide: https://api.agentmailer.ai/llms.txt
• AgentMailer authentication guide: https://api.agentmailer.ai/auth.md
• Documentation: https://agentmailer.ai/docs
• Support: support@agentmailer.ai

Triage order

1. Confirm the configured URL is exactly the HTTPS MCP endpoint above.
2. Confirm the client supports remote Streamable HTTP and OAuth discovery.
3. Reconnect or clear only the AgentMailer connection when authorization is stale. Do not delete unrelated client credentials.
4. Call auth_me. A success proves authentication and reports the organization and permission set.
5. Confirm auth_me reports a trusted identity before inbox creation. If not, complete human approval using the existing approval URL or the signup sequence in llms.txt.
6. Call list_inboxes. An empty result is valid and is not a transport failure.
7. If a tool returns forbidden, compare the requested operation with the permissions returned by auth_me; do not retry with guessed organization or resource identifiers.

Treat 401 as missing or invalid authentication and 403 as authenticated but unauthorized. human_approval_required specifically means the approval ceremony is incomplete. Do not expose access tokens, authorization headers, approval codes, or reviewer credentials in diagnostics.