MCP payload conventions¶

Payload conventions¶

Short reference for response structure patterns across the tool surface.

IDs — Run IDs are 8-char hex handles. Finding IDs are short prefixed forms. Both accept the full canonical form as input.

Detail levels — summary (default for lists), normal (default for single finding), full (compatibility payload with URIs).

Pagination — list_findings and get_report_section(section="metrics_detail") support offset and limit. list_hotspots supports limit and max_results only (no offset).

Changed-scope filters — list_findings, list_hotspots, and generate_pr_summary accept changed_paths or git_diff_ref for PR projection.

Threshold context — Empty check_* responses include threshold_context showing whether the run is genuinely quiet or simply below the active threshold.

Budget nulls — check_patch_contract uses null for disabled numeric thresholds. Boolean policy gates use forbid_* names.

Long context — do_not_touch, review_context, and similar sections include total, shown, and truncated summaries.