Coroner provides 7 tools with explicit schemas and descriptions visible in MCPEngine.swift and Digest.swift. All tools have descriptions (10-150 chars) and input schemas with typed parameters. However, descriptions are terse and lack LLM-optimized guidance on WHEN to use each tool or HOW they chain together. Parameter descriptions are minimal or absent (e.g., 'id' in crash_detail has no description). No output schemas documented. Error handling and recovery guidance are absent. Tool naming is reasonable (verb_noun style: new_since, top_crashes, crash_detail) but some names are domain-specific jargon ('suspects', 'digest') that may confuse LLMs unfamiliar with post-mortem analysis. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all tools being read-only. Composition is sound, tools are single-purpose and chain logically (new_since → top_crashes → crash_detail → suspects). Missing: parameter constraints (enums for 'kind' are present, but 'period' enum values lack descriptions), output field documentation, and error recovery patterns.

• Add 1-2 sentence descriptions to every parameter. E.g., 'id: cluster identifier returned by top_crashes or new_since' for crash_detail.
• Document output schemas for each tool. E.g., 'Returns: {cluster_id: string, signature: string, frame_count: number, first_seen_build: string, occurrences: number}'.
• Add readOnlyHint: true to all 7 tool definitions to signal safety to LLMs.
• Expand tool descriptions to include WHEN to use and HOW they chain. E.g., 'Call new_since first after a release to find new failures, then top_crashes to rank by frequency, then crash_detail to inspect individual clusters.'
• Add error recovery guidance. E.g., 'If cluster id not found, try top_crashes to list valid ids' for crash_detail.