Github Copilot App cannot run MCP tools anymore

I'm currently unable to run Kusto through MCP from Github Copilot App. It started on September 10th and continues the next day. I tried to restart Github Copilot App, test in different sessions, to no avail.

The failure seems at the Copilot App extension/tool registry layer, not Kusto authorization or my query. In the affected sessions, the Kusto server had previously been visible, then list_resources(["kusto"]) returned no resources; before that, its query executor was absent even when discovery functions remained. The local fallback also could not be used because Kusto.Cli.exe was not installed and package retrieval failed due to the local NuGet/TLS configuration.

Reproduction:

1. Open any existing or newly forked project session.
2. Ask the agent to perform a Kusto query.
3. The agent’s tool registry reports kusto as deferred/unavailable, and discovery resolves to no callable kusto_query.
4. The same user/session previously had successful Kusto MCP calls, so this is a regression in extension loading or capability registration—not a cluster permission failure.
5. Impact: prevents production telemetry analysis and transcript correlation; the app cannot complete routine engineering investigation work.

Approximate onset: 2026-09-10/11, and examples of impacted sessions: Office 365 error triage fork (cc0bcb3b-ec11-4c36-91e7-07b65244b1f7) and the originating Office 365 triage session (c1bbf79b-ebc4-4eb5-8684-c2ccb3332088).

No source files or tests are named. Start by reproducing the failure in an existing or newly forked project session using a Kusto query, then inspect the app's extension/tool registry behavior around deferred or unavailable tools. Done means the Kusto MCP server and query executor are discoverable and callable again in the affected sessions.