MCP tooling infrastructure: client, code generation, tool server, and approval workflow
Four tools with mixed quality. tool-1, tool_2, and tool_3 have minimal descriptions (10-50 chars, well below the 194-char baseline) and lack parameter descriptions for most inputs. The 'fetch' tool is better documented but still has gaps. No tool annotations (readOnlyHint/destructiveHint) despite all being READ_ONLY. Naming is inconsistent (tool-1 vs tool_2 vs tool_3 vs fetch). Schemas present but sparse, tool_2 has a null description for 'delay' parameter. No output schemas documented. Error handling and recovery guidance absent.
Fetches a URL from the internet and extracts its contents as markdown.
This is tool 1.
This is tool 2.
This is tool 3 with nested structured output.
Three test tools (tool-1, tool_2, tool_3) have trivial descriptions ('This is tool 1.', 'This is tool 2.', 'This is tool 3 with nested structured output.'), 10-50 chars, far below the 194-char baseline. LLMs cannot determine when or why to select these tools.
tool_2 has a parameter 'delay' with null description. All parameters must have non-empty descriptions explaining what they control. LLMs cannot infer parameter meaning from names alone.
No output schemas documented for any tool. LLMs need to know what fields to expect so they can plan downstream tool calls and extract the right data.
Tool naming is inconsistent and non-descriptive. 'tool-1', 'tool_2', 'tool_3' do not start with action verbs and do not convey what happens when called. 'fetch' is the only verb-based name. Inconsistent naming (hyphens vs underscores) adds friction.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 47 | 2025-06-18+ | v2 |
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all tools being READ_ONLY. Tool annotations are part of the current MCP spec (2026-07-28) and help LLMs reason about safety and retry behavior.