hayate-mcp is a conformance test fixture, not a production tool server. 27 tools are defined with schemas and descriptions present. However, most tools have minimal, generic descriptions (10-30 chars: 'Tests X content', 'Exercises SEP-2322'). Parameter descriptions are absent, tools declare input schemas but do not annotate individual parameters. Two 'echo' tools are duplicates (one with additionalProperties=false). Most tools accept empty schemas (no parameters), limiting composability. Error handling is present but not documented in tool descriptions. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all tools being READ_ONLY. Naming is clear and verb-forward (test_*, json_schema_*), but descriptions do not explain WHEN to call each tool or what downstream actions they enable.

• Expand tool descriptions to 50 - 200 characters, explaining WHAT the tool does, WHEN to call it, and what it returns. Example: 'test_simple_text' → 'Validates server ability to return plain text content blocks. Call to verify text response formatting and encoding.'
• Add individual parameter descriptions to all tools with inputs. For 'test_sampling', annotate 'prompt' as 'The prompt string to send to the LLM for sampling (required, non-empty).'
• Remove or consolidate duplicate 'echo' tools. If both are needed, rename one (e.g., 'echo_strict') and document the difference in descriptions.
• Add tool annotations: set readOnlyHint=true on all 27 tools to signal they are safe for agents to call without side effects.
• Document output schemas in tool descriptions or via a separate schema field. Example: 'Returns a ContentBlock array with text, image, audio, or embedded resource content.'
• For MRTR tools (test_input_required_result_*), clarify in descriptions that they exercise multi-round-trip input flows and what state/input they expect across rounds.

Inferred effective spec: 2026-07-28+.

• Relies on Sampling (deprecated) - integrate directly with the LLM provider API