Single tool with complete schema and good description. Tool name uses verb_noun pattern (migrate_feishu_wiki). Input schema properly typed with required fields. Description explains WHAT (migrate docs), WHEN (user logged in), and includes dry_run safety mechanism. However, output schema is undocumented, LLM cannot predict response structure. Error handling returns JSON but lacks recovery guidance. Tool annotations present (destructive=True, idempotent=False) which is excellent. Missing: parameter-level descriptions for output fields, explicit error classification, and guidance on what to do if migration fails.

• Document the output schema in mcp_server.py. Add a comment or docstring showing expected response structure: {success: bool, migrated_count: int, failed_count: int, errors: [{item, reason}], dry_run_mode: bool}.
• Enhance error messages with recovery guidance. Catch specific exceptions (FileNotFoundError, PermissionError, JSONDecodeError) and return actionable messages like 'lark-cli not found. Install with: pip install lark-cli' or 'Authentication failed. Run: lark-cli auth login'.
• Add format constraints to parameter descriptions. Update 'source' description to: 'Source Feishu wiki URL (https://...) or node token (e.g. docx_abc123def456). Required.' Similarly for 'target'.
• Validate input format before execution. Add regex or URL parsing to reject obviously invalid source/target values early with clear error messages.
• Include dry_run_mode flag in response. Ensure JSON response always includes {dry_run_mode: true/false} so LLM knows whether results are simulated or committed.