[Enhancement]: User self-diagnosis for MCP failures without backend log access

[peeeteeer]

What features would you like to see added?

As we started rolling out the mcp feature UI we do realize that a lot of the integrations fail and do need access to the log to find errors. With large deployments of multiple thousands of users that becomes in a support nightmare. Therefore we would love to see...

a) a more sophisticated error handling and display at the UI especially in the "create mcp server" dialog
b) somehow an access to log/telemetry data per mcp server

More details

some initial ideas...

1. Status API enhancement
   Update GET /api/mcp/connection/status payload from only state to state + reason.

Proposed MCPServerStatus:

• connectionState
• requiresOAuth
• lastError?: { code: string; message: string; retryable: boolean; traceId?: string; at: string }

2. Normalize MCP error codes
   Right now error strings are inconsistent and often generic. Extend the existing error codes with stable codes and keep message user-safe such as

MCP_AUTH_REQUIRED
MCP_OAUTH_TIMEOUT
MCP_DNS_FAILURE
MCP_CONNECTION_REFUSED
MCP_PROXY_SSE_ISSUE
MCP_DOMAIN_NOT_ALLOWED
MCP_INSPECTION_FAILED
MCP_INTERNAL_ERROR

3. UI toast/status mapping
   Replace generic com_ui_mcp_init_failed with code-aware messaging and call to actions such as

MCP_AUTH_REQUIRED -> “Authentication required for {server}” + open OAuth.
MCP_DNS_FAILURE -> “Server URL cannot be resolved” + open Edit dialog.
MCP_PROXY_SSE_ISSUE -> “Connection blocked by proxy/SSE config” + docs link.
MCP_INTERNAL_ERROR -> “Server error, try again” + show traceId.

4. Trace ID for support
   For each MCP error response, include traceId (UUID).

Log traceId server-side with the full error.
Show small “Debug ID: …” in toast/modal so users can report it.

and after that ....

1.) Emit structured diagnostics events at MCP touchpoints (connect, OAuth, tool call, transport error).
Include McpServerName, userId, conversationId/runId, code, severity, messageSafe, traceId.

2.) Store in a queryable TTL store (Mongo capped/TTL collection or Redis stream).
Retention 7–14 days, indexed by (serverName, userId, timestamp).

3.) Add scoped endpoint:
GET /api/mcp/:serverName/diagnostics?cursor=&severity=&from=&to=
RBAC: only users with access to that mcp server (plus admin override).

4.) UI: add “Diagnostics” drawer on MCP server card/menu showing timeline and “Copy Debug ID”.

Which components are impacted by your request?

No response

Pictures

No response

Code of Conduct

• I agree to follow this project's Code of Conduct