The News, Explained
GitHub announced on October 22, 2026 that Copilot usage reports now identify custom skills, agents and MCP servers by their configured names. The API lets software retrieve those reports. Administrators gain a clearer view of their own tools, but each number needs to be read against the event it measures. Source
A skill supplies instructions for a task; a custom agent combines a role with tools. MCP is a protocol connecting AI to external tools. The reports count skill invocations, agent starts and MCP connection or reconnection attempts. Further tool calls over an existing MCP connection do not increase its count. Plugin-associated skill invocations appear in both skill and plugin totals, so adding those totals double-counts activity. Source
When names appear
Names appear in per-user 1-day and 28-day reports and enterprise or organization reports. Daily reports and daily records within aggregated 28-day reports retain up to 25 named entries per customization type, ordered by recorded events. Remaining entries are grouped into other, which can appear in addition to the 25 names. Absence from a capped list does not establish zero use. Source
Access requires the Copilot usage metrics policy and an eligible role: enterprise owner or billing manager, organization owner, or a custom role with metrics-viewing permission. Names come from Copilot CLI interactive and prompt modes, the Copilot app, ACP integrations and the Copilot SDK. CLI is the command-line tool, ACP connects agents with clients such as editors, and an SDK is a development kit for embedding Copilot in an application. Source Source
Prompt mode, the app and ACP integrations require CLI 1.0.97 or later; the SDK requires 1.0.21 or later. An overridden SDK-bundled CLI must also be at least 1.0.97. Built-in IDE Copilot chat integrations do not report names. Configured names are trimmed to 128 characters; path-like or unavailable names remain grouped. Review those names before sharing exports outside the organization. Names appear for report days processed after deployment; earlier days are not automatically backfilled. Source
Events, identifiers and people
interaction_count measures the events above. The distinct fields count different item identifiers used: several people using one skill still contribute one identifier to the organization-wide count. Adding per-user distinct values therefore need not reproduce the aggregate. Outdated clients or blocked telemetry can leave gaps, limiting exact accounting and complete audits. Source
OYOPICK’s Take
Seeing an organization’s tools by name is useful progress. Yet an agent start and an MCP connection capture different stages of work. A larger number alone cannot establish wider adoption or better results.
For deciding where enablement is needed, we would compare the same kind of item under consistent client and reporting-period conditions. Historical names are not backfilled, so a before-and-after trend needs that boundary accounted for. The stronger opportunity is to connect named usage with why a tool helped people or where they got stuck, so organizations can improve their tools instead of optimizing a counter.