A Mastra-based agentic system with multiple agents, workflows, and tools for weather, financial analysis, content processing, learning assistance, and personal assistance tasks.
This server has 2 tools with functional schemas and descriptions, but significant quality gaps prevent a higher score. Both tools have verb-based names (get-transactions, get-weather) and documented output schemas. However, parameter descriptions are sparse or missing, error handling is absent, and neither tool includes critical metadata for LLM safety and composition. The get-transactions tool has an empty input schema with no description of why no parameters are needed. The get-weather tool has one parameter with a minimal description. Neither tool declares their behavior (idempotent, read-only enforcement, timeout handling). Output schemas are present but lack detailed field descriptions. No error recovery guidance, no rate limiting, no input validation messaging. The server exposes external API calls (Google Sheets, Open-Meteo, Geocoding) without documented timeout behavior, error cases, or fallback strategies.
Get transaction data from Google Sheets
Get current weather for a location
get-transactions has an empty input schema (no parameters) but lacks documentation explaining why, LLMs cannot infer whether the tool requires context, state, or authentication to function. Per Dimension 1.C, every parameter needs a description; when a tool takes NO parameters, the description must explain this design decision.
get-weather parameter 'location' has a minimal description ('City name') but lacks guidance on format, expected precision, or handling of ambiguous names (e.g., 'Springfield' matches 20+ US cities). Per Dimension 1.C baseline, average param annotation is 72 chars; this is ~9 chars. No enum, no constraint, no dependency hints.
Neither tool includes error handling guidance. get-transactions fetches from a public Google Sheet, what if the sheet is deleted, access revoked, or the API times out? get-weather calls two external APIs in sequence, no timeout documentation, no fallback if geocoding fails or returns ambiguous results. Per Dimension 1.E, error responses must tell the LLM what to do next.
Inferred effective spec: 2025-06-18+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 44 | 2025-06-18+ | v2 |
| 2026-03-09 | F | 27 | - | v1 |
Tool definitions lack annotation hints (readOnlyHint, destructiveHint, idempotentHint) required by current MCP spec (2026-07-28). Both tools are read-only and idempotent, but this is not declared, agents cannot infer safety properties from names alone.
Output schemas for both tools lack detailed field descriptions. get-transactions returns csvData (string) with no hint that it's raw CSV format, no encoding declaration, no hint on how to parse or validate. get-weather returns 7 fields with no descriptions, agents must infer that 'temperature' is numeric, 'conditions' is a weather code string, etc.
get-transactions description ('Get transaction data from Google Sheets') does not state WHEN to use it (e.g., before calling the financial agent), what it returns (CSV format), or any prerequisites (e.g., the Sheet URL is hardcoded, user cannot supply their own). Per Dimension 1.B baseline, average tool description is 194 chars; this is ~45 chars.
get-weather calls external APIs (open-meteo geocoding and weather) with no documented timeout, retry logic, or fallback. If geocoding returns 0 results, the execute function throws a generic Error, no guidance to the LLM on alternative queries (e.g., 'Try a country name or broader region'). Per Dimension 1.E, errors must guide recovery.
No input validation or constraint enforcement visible. get-weather accepts 'location' as a free-form string, LLMs may pass special characters, Unicode variants, or SQL-like injection attempts. No min/max length, no regex, no enum. Per Dimension 1.C, constraints prevent hallucinated invalid values.