A first model context protocol server that implements a simple notes system demonstrating core MCP concepts like resources and tools by allowing listing notes as resources, reading individual notes, creating new notes via a tool, and summarizing all notes via a prompt
Two tools are present with complete input schemas and descriptions. Naming follows verb_noun convention (create_note, calculate). Descriptions are present but brief (27-88 chars vs. baseline 194 chars). Schemas are properly typed with required fields. However, descriptions lack depth on WHEN to use each tool, what they return, and error recovery paths. No output schemas are documented. Error handling is minimal, validation errors are generic (e.g., 'Title and content are required') without recovery guidance. Tool composition is reasonable (two independent responsibilities), but parameter descriptions could be more prescriptive about constraints and formats.
Perform simple calculations (add, subtract, multiply, divide)
Create a new note
Tool descriptions are too brief (27-88 chars vs. baseline 194). 'Create a new note' and 'Perform simple calculations...' lack context on WHEN to use the tool, what it returns, and error conditions. LLMs cannot distinguish this tool from similar ones or plan downstream calls effectively.
Output schemas are not documented. Callers cannot plan what fields to expect from create_note (does it return note_id, uri, timestamp?) or calculate (does it return result, unit, precision?). This forces LLMs to infer structure or assume, risking extraction errors.
Parameter 'operation' in calculate uses an enum (add, subtract, multiply, divide), which is good. However, the enum is not reflected in the description text, LLMs cannot read JSON Schema enums directly. Add to description: '(one of: add, subtract, multiply, divide)'.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 52 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 40 | - | v1 |
Error handling is generic. Validation errors (e.g., 'Title and content are required', 'Unknown operation') do not tell the LLM what to do next. No recovery guidance (e.g., 'Did you provide both title and content?', 'Valid operations: add, subtract, multiply, divide').
The create_note tool modifies state (persists to in-memory notes dictionary) but the description does not state this. Agents need to know which calls have irreversible consequences. Add: 'Creates a persistent note. This action cannot be undone.'