PolyForge MCP exposes 4 domain-specific tools for evidence ledger management with clear, verb-prefixed names (evidence_append, evidence_verify, gate_evaluate, gate_report). All tools have descriptions (50-150 chars) and input schemas with typed parameters. However, schemas lack output documentation, parameter descriptions are minimal (10-30 chars), and error handling guidance is absent. No tool annotations (readOnlyHint/destructiveHint) despite clear risk profiles. Naming is strong; schema completeness and error recovery are weak.

• Document output schemas for all tools. E.g., gate_report should return {entries: [{id, kind, state, ...}], bundle_sha256: string, gate_pass: boolean}.
• Expand parameter descriptions to 50-100 chars with format/constraint hints. E.g., task_id: 'Unique task identifier (alphanumeric, 1-64 chars, immutable)'.
• Add tool annotations: evidence_append/evidence_verify with destructiveHint=true; gate_evaluate/gate_report with readOnlyHint=true.
• Include error recovery guidance in tool descriptions. E.g., evidence_verify: 'If tool_name is not allowlisted, call list_allowlisted_tools() to see available options.'
• Define eval_metadata schema as a typed object with documented fields (e.g., {experiment_id?: string, run_id?: string, ...}) or accept a JSON string with validation.