Tools & Usage

What the Tools page shows

Navigate to Tools & Usage in the sidebar to see aggregated data about every tool call captured by the Trenchcoat plugin. This page answers questions like:

• Which tools does my agent call most often?
• Are any tools failing at an unusual rate?
• Is a particular tool getting slower over time?

What counts as a "tool"?

Trenchcoat tracks any tool that appears in a tool_use event. In Claude Code this includes:

Built-in Claude Code tools:

• Read — read file contents
• Write — write or create a file
• Edit — make targeted edits to a file
• Bash — run a shell command
• Glob — find files by pattern
• Grep — search file contents
• WebFetch — fetch a URL
• WebSearch — run a web search

MCP tools: Any tool exposed by a connected MCP server appears here by its fully qualified name (e.g., mcp__supabase__query).

Subagent tools: Tool calls made by spawned subagents are also captured if the plugin is active in the subagent process.

Top tools table

The main table on the Tools page lists every tool that was invoked at least once during the selected date range, ranked by total invocations descending.

How error rate is calculated

Error rate is computed as:

error_rate = tool_result events where is_error = true
             ─────────────────────────────────────────
             total tool_use events for that tool

A tool_result is marked as an error when Claude Code sets is_error: true on the result — this happens for things like file-not-found, permission denied, non-zero Bash exit codes, and network failures in WebFetch.

Usage over time chart

Below the table, a line chart shows daily invocation counts for the top 5 tools (by total volume) over the selected date range. Each tool gets its own colored line. Hover over any point to see exact counts for that day.

This chart is useful for spotting correlated spikes — if Bash and Write both spike on the same day, that likely maps to a big coding session.

Filtering

• Date range — select the time window using the date range picker (top right). The table and chart both update immediately.
• Minimum invocations — enter a threshold to hide low-volume tools. This reduces noise when you have many rarely-used MCP tools and want to focus on the ones that matter.

Clicking a tool row

Click any row in the top tools table to see a detail view for that tool. The detail view shows:

• Invocations per day (chart)
• Error rate over time (chart)
• A sample of recent tool_use + tool_result event pairs, with expandable payloads

This is useful for debugging a specific tool — you can inspect the exact inputs that led to an error.