Two tools with complete input schemas and Japanese descriptions, but significant gaps in LLM-optimization and error guidance. Tool names lack action verbs (neighbors, subgraph are nouns). Descriptions are domain-specific but lack WHEN/WHY context for LLM selection. No output schema documentation. Parameter descriptions are minimal (駅名, 中心の駅名). Error handling returns text blobs without structured guidance. No tool annotations (readOnlyHint present in metadata but not declared in schema). Baseline: avg tool description ~194 chars; these are ~120 chars. Param descriptions avg 72 chars; these are ~20 chars.

• Rename tools to action verbs: 'get_adjacent_stations' (or 'list_neighbors') and 'get_station_subgraph'. This lets LLMs infer intent from the name alone.
• Expand parameter descriptions to 50-100 chars each. For 'station': 'Station name in kanji (e.g. 渋谷) or romaji (e.g. Shibuya). Suffix 駅 is optional. If ambiguous, returns candidates.' For 'radius': 'Number of hops from center station (1-4, default 2). Larger radius returns more edges but slower.'
• Document output schema in tool description. E.g. 'Returns formatted text with: (1) center station name, (2) list of adjacent stations with line names, (3) transfer stations, (4) lines passing through center.' Mention that output is human-readable text, not JSON.
• Add structured error responses. When station is ambiguous, return JSON with 'error', 'candidates' array, and 'suggestion' field. E.g. {"error": "ambiguous_station", "candidates": ["渋谷 (JR)", "渋谷 (メトロ)"], "suggestion": "Try 'get_adjacent_stations(station=\"渋谷 (JR)\")'"} instead of plain text.
• Add readOnlyHint: true to both tool schemas in list_tools(). This signals to LLMs that these tools are safe to call without confirmation and can be retried.