A demonstration MCP server with tools for sticky notes and weather forecasts
This server has 4 tools with basic definitions but significant quality gaps. Naming is reasonable (verb_noun pattern), but descriptions lack depth and LLM-specific guidance. Input schemas are present but lack important constraints (enums, ranges, formats). Output schemas are entirely undocumented. Error handling is minimal, tools return generic strings with no recovery guidance. No tool annotations (readOnlyHint/destructiveHint/idempotentHint) despite clear risk differences. The `add_note` tool is particularly concerning: it mutates state (WRITE risk) but has no confirmation pattern, idempotence guidance, or error classification. Tools are reasonably well-composed (each does one thing), but parameter descriptions are sparse. For a sticky notes and weather demo, this is acceptable, but for production use, substantial improvements are needed.
Append a new note to the sticky note file.
Get weather alerts for a US state.
Get weather forecast for a location.
Read and return all notes from the sticky note file.
Destructive tool (add_note) lacks confirmation pattern and state-mutation guidance. LLMs don't know this call has irreversible consequences.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk differentiation. add_note is WRITE, others are READ_ONLY, but LLMs see no metadata distinguishing safety profiles.
Output schemas completely undocumented. Tools return plain strings or dicts but LLMs have no specification of return type, structure, or fields for downstream composition.
Parameter descriptions lack actionable constraints. 'state' example 'e.g. CA, NY' invites literal reuse rather than proper enum. Latitude/longitude lack range constraints (-90 to 90, -180 to 180).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 46 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 6 | - | v1 |
Error handling provides no recovery guidance. Tools return generic strings ('Unable to fetch...'). LLMs don't know if the error is retryable, user-fixable, or fatal.
Tool descriptions are too short (10-60 chars) and lack LLM-specific guidance on WHEN to use each tool, dependencies, and typical next steps.