Which built-in agent tools get used most?

Call volume captures repetition; developer reach captures how broadly an exact tool name appears. The charts keep those signals separate because a frequently looped shell action and a broadly available read action describe different behavior.

Calls, reach, errors, and calling-turn context

The table is ordered by calls and limited to the top 25 tools meeting the comparison floor. Identifiers remain exact: similarly named actions from different agents are not merged without evidence that they represent the same operation.

What each tool-usage metric means

Calls

The number of observed invocations for one exact built-in tool identifier in participating developers' latest submitted transcript snapshots. MCP names beginning with mcp__ are excluded from this page.

Developer reach

Distinct listed developers whose latest snapshot includes that identifier. One developer can use many tools, so row counts overlap and must not be added into a cohort total.

Error rate

Reported error outcomes divided by calls for that exact identifier. It is an operational signal, not a quality score: invalid commands, permission issues, expected retries, and tool failures can all contribute.

Turn-attributed tokens

Tokens on the model turn that invoked a tool, split across multiple tool calls on the same turn. They measure surrounding model context—not tokens consumed by the tool itself and not a direct execution fee.