22 tools with consistent naming (verb_noun pattern: hedera_get_*, hedera_create_*, hedera_transfer_*, hedera_*). All tools have descriptions (avg ~80 chars, within baseline 34-392). Input schemas present for all tools with typed parameters and descriptions. However, output schemas are NOT documented, responses are free-text serialized transactions or JSON strings without declared structure. Error handling returns generic 'Error: {message}' without recovery guidance or categorization. No tool annotations (readOnlyHint/destructiveHint/idempotentHint) despite clear risk levels (READ_ONLY, WRITE, IRREVERSIBLE). Parameter descriptions are minimal (e.g., 'Token ID' vs 'Token ID (format: 0.0.X where X is numeric)').

• Document output schemas for all tools. Specify: for create_* tools, return {transactionBase64: string, transactionId?: string, ...}; for get_* tools, return {rate: number, timestamp: string} or {blocks: [{blockNumber: number, timestamp: string, ...}], total: number}. This enables LLM chaining and context preservation.
• Add tool annotations to index.ts register() calls. Use server.tool(..., {readOnlyHint: true}) for hedera_get_*, {destructiveHint: true} for hedera_burn_*, {idempotentHint: true} for hedera_decode_*. Requires MCP SDK ≥1.13.0.
• Enhance error handling in register() handler. Catch errors and return structured responses: {isError: true, content: [{type: 'text', text: 'Invalid tokenId format. Expected 0.0.X (e.g., 0.0.123). Got: {value}. Retry with valid format.'}]}. Include recovery hints for common failures (not found, invalid format, permission denied).
• Expand parameter descriptions with format constraints. E.g., 'Token ID (format: 0.0.X where X is numeric, e.g., 0.0.123)' instead of 'Token ID'. Add ranges for numeric params: 'Limit (1-100, default 20)'.
• Add optional dry-run parameter to irreversible tools (hedera_burn_token, hedera_freeze_token_account). When dry_run=true, return 'Would burn X tokens from {tokenId}' without executing. Prevents accidental destruction.