Production-ready MCP server starter — SDK 1.29, 3 example tools, HTTP/SSE transport. Fork to build your own Claude tools.
Three tools with complete schemas and reasonable descriptions. Naming follows verb_noun convention (search_, generate_, analyze_). Descriptions are adequate (60-120 chars) but lack actionable context for LLM selection. Parameters are typed and constrained (enums, limits), but descriptions are minimal. No output schema documentation. Error handling returns text but lacks recovery guidance. Missing tool annotations (readOnlyHint). Composition is sound, each tool has one responsibility.
Analyze text and return stats — word count, character count, reading time, sentiment hint, and top tokens.
Generate a short mythic-voiced description for a subject. Useful for worldbuilding, fiction, and creative writing.
Search the local knowledge base for entries matching a query. Returns ranked results with excerpts.
Output schemas not documented. LLMs cannot plan downstream calls or extract fields without knowing what analyze_text, generate_lore, and search_knowledge return.
Parameter descriptions are minimal. 'Search terms' and 'What to describe' lack context for LLM selection. Should explain when to use each tool vs alternatives, expected input format, and dependencies.
Tool descriptions lack WHEN/WHY context. 'Search the local knowledge base' does not explain when to call this vs a web search or database query. LLMs need explicit selection guidance.
Error responses return plain text without recovery guidance. 'Tool error: <message>' gives LLM no actionable next step. Should categorize as retryable/user-fixable/fatal and suggest alternatives.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
No tool annotations (readOnlyHint). All three tools are read-only but lack explicit destructiveHint=false or readOnlyHint=true. Clients cannot distinguish safe tools from destructive ones.