Named Copilot Usage Metrics Enable Precise Adoption Tracking

GitHub Copilot’s usage metrics API now surfaces the explicit names of custom skills, agents, Model Context Protocol (MCP) servers, slash commands, and plugins rather than aggregating them under generic "other" or "custom" categories. The change also expands each totals_by_* list from five to up to twenty‑five entries, exposing more of the long tail of internal customizations. Practitioners care because the data now reflects the real adoption of organization‑specific extensions, enabling targeted enablement and more accurate reporting.

What Changed in the API

The following arrays now contain named entries:

• totals_by_skill
• totals_by_custom_agent
• totals_by_mcp
• totals_by_slash_cmd
• totals_by_plugin

Each array can list up to twenty‑five distinct names per record; any remaining activity is collapsed into a trailing "other" entry. Interaction counts and distinct use counts remain calculated as before.

Why This Matters for Engineers

Visibility into named customizations lets platform and DevOps teams answer questions such as:

• Which internally built skills are actually invoked by developers?
• Is a newly rolled‑out custom agent gaining traction?
• Are MCP servers being used consistently across teams?

These insights support capacity planning, documentation updates, and focused training. Security engineers can also spot unexpected or orphaned custom components that may need review.

Operational and Security Implications

To receive named data, telemetry must originate from supported clients:

• Copilot CLI (interactive sessions and prompt mode) version 1.0.97 or later.
• Copilot SDK version 1.0.21 or later (including JetBrains ACP integrations).
• IDE integrations that run on the supported SDK version.

Names are reported exactly as configured, trimmed to 128 characters, and file‑path‑like strings remain grouped under "other". Because the metrics rely on client telemetry, outdated clients or blocked telemetry can cause under‑counting. Practitioners should verify that the Copilot usage metrics policy is enabled and that the appropriate "View Copilot Metrics" permission is granted to the roles that need the data.

Related CloudNinjas coverage: AI engineering.

What This Means For Practitioners

Adopt the following actions:

1. Confirm all developer workstations run the required CLI/SDK versions to avoid missing data.
2. Review naming conventions to keep identifiers under 128 characters and avoid path‑like patterns if you need them reported.
3. Update monitoring dashboards to consume the expanded totals_by_* arrays and adjust alert thresholds based on the longer tail of entries.
4. Periodically audit telemetry health—ensure no firewall or policy blocks the Copilot client from sending usage events.
5. Use the new granularity to drive adoption campaigns for high‑value custom components and to retire unused ones.

By integrating these named metrics into existing observability pipelines, teams gain a clearer picture of internal Copilot extensions and can make data‑driven decisions about rollout, support, and security review.