Single tool with partial schema and Portuguese description. Input schema present with types and descriptions (dashboard_uid: string, panel_id: integer), meeting baseline requirements. However, description is in Portuguese, limiting LLM comprehension in English-first environments. No output schema documented, LLMs cannot predict return structure. Tool name 'get_panel_data' follows verb_noun convention correctly. No error handling guidance, no parameter constraints (e.g., panel_id range), no idempotency hints. Risk classification (READ_ONLY) present but not integrated into description. Missing: pagination guidance, result limits, field documentation for complex nested response.

• Add English-language description (50-200 chars) explaining what the tool does, when to use it, and what it returns. Example: 'Fetch time-series data from a Grafana panel. Automatically discovers datasource and query from dashboard config. Returns structured data with dashboard/panel metadata and time-series results.'
• Document output schema with field types and descriptions. Specify that 'data' contains array of series objects with fields like refId, values, timestamps. Include example structure.
• Add error recovery hints to description: 'If panel not found, call list_panels(dashboard_uid) first. If panel is not timeseries type, use get_panel_config() to inspect available panel types.'
• Constrain panel_id parameter: add description detail 'Numeric panel ID (typically 1-999). Find valid IDs by inspecting dashboard JSON or calling list_panels().'
• Add result limit guidance: 'Returns up to 1000 data points per query. For longer time ranges, use narrower time windows or increase maxDataPoints parameter in dashboard config.'