Add MCP tool usage statistics to logs command

[Copilot]

Extends gh aw logs to display aggregated MCP tool call statistics from gateway.jsonl files, reusing the data structures introduced in #13067 for the audit command.

Changes

Data structures

• Added MCPToolUsageSummary to LogsData for aggregating tool usage across runs
• Extended ProcessedRun, RunSummary, and DownloadResult with MCPToolUsage field

Data extraction and aggregation

• Calls extractMCPToolUsageData() for each run in logs_orchestrator.go
• Implemented buildMCPToolUsageSummary() to merge statistics across runs:
  • Weighted averaging for duration metrics
  • Max tracking for I/O sizes
  • Aggregates server stats and per-tool summaries

Output

• Console renders two tables: server statistics (requests, tool calls, I/O, duration, errors) and tool statistics (calls, I/O sizes)
• JSON includes complete tool call history
• Individual tool calls excluded from console display

Example

gh aw logs my-workflow --json

Returns aggregated MCP tool usage in mcp_tool_usage field with server stats, tool summaries, and individual call records for analysis.

Original prompt

Reuse the data structures for MCP tool analysis in #13067 and apply it in "logs" so that user can see summarized tool call information.

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.

[pelikhan]

@copilot runs logs command and show output

[Copilot]

@copilot runs logs command and show output

The gh aw logs command now displays MCP tool usage statistics in both console and JSON formats.

Console Output:

The command shows two new tables when MCP servers are detected in workflow runs:

1. Server Statistics - Aggregated metrics per MCP server:

   • Request count and tool call count
   • Total input/output sizes
   • Average duration across all calls

2. Tool Statistics - Detailed per-tool breakdown:

   • Call count per tool
   • Total and max I/O sizes
   • Performance metrics

Example Console Output:

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
                            🔧 MCP Tool Usage
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

                            Server Statistics
────────────────────────────────────────────────────────────────────────────────────────

Server       Requests  Tool Calls  Total Input  Total Output  Avg Duration
github       8         8           4.5 KB       18.2 KB       145ms
playwright   4         4           1.8 KB       3.5 KB        165ms

                             Tool Statistics
────────────────────────────────────────────────────────────────────────────────────────

Server      Tool             Calls  Total Input  Total Output  Max Input  Max Output
github      search_issues    5      3.2 KB       15.1 KB       1.5 KB     6.0 KB
github      get_repository   3      1.3 KB       3.1 KB        512 B      2.0 KB
playwright  navigate         2      1.0 KB       2.0 KB        512 B      1.0 KB
playwright  click            2      800 B        1.5 KB        400 B      800 B

JSON Output:

With --json flag, the output includes mcp_tool_usage field with:

• summary: Aggregated per-tool statistics
• servers: Server-level statistics
• tool_calls: Complete history of individual tool calls with timestamps

The data is aggregated across all downloaded workflow runs, with weighted averaging for duration metrics and max tracking for I/O sizes.

See full demo output: /tmp/demo_output.md