On September 17, 2026, GitHub announced Copilot CLI customization metrics for skills, custom agents, MCP servers, slash commands and plugins. The update extends CLI usage reporting with visibility into which customizations people use. Read GitHub’s announcement.
What the new metrics show
Reports include up to five leading items per category, interaction counts and distinct-item counts. Customer-defined names are grouped for privacy. MCP interactions count connection and reconnection attempts, including failures; repeated tool calls on an existing connection do not increase that count. Plugin skill invocations also appear in skill totals, so adding those totals would double-count activity.
Zero counts or empty arrays mean no matching activity; null or missing fields mean data is unavailable. Access requires an eligible administrative role or metrics permission, plus an enabled usage metrics policy.
The official API documentation describes daily report downloads through signed URLs that expire. Reporting pipelines should retrieve and store the downloaded report rather than treating its download URL as a permanent source.
Why this matters for QA engineers
QATechTools analysis: These signals can help a QA lead track adoption of assistant customizations, but they do not demonstrate that a regression suite passed or a defect was fixed. A spike in MCP activity could warrant investigation of reconnect behavior before being interpreted as increased productive use.
For teams importing the data into dashboards, useful checks include preserving the difference between missing and zero values, avoiding overlapping skill/plugin totals, and displaying connection attempts separately from test outcomes. Pair adoption reporting with your own CI results, reviewed changes and defect evidence before drawing conclusions about quality.
Verified September 18, 2026. The QA recommendations above are editorial analysis, separate from GitHub’s announced functionality.