A simple TypeScript MCP server for demo and presentation purposes
This server exhibits foundational issues across naming, descriptions, and schema design. All 6 tools have zero-parameter schemas (empty properties objects), violating the principle that every tool should accept at least implicit context or options. While tool names follow verb_noun convention (what_is_*, get_*), descriptions are generic educational summaries that do not guide LLM selection or explain WHEN to call each tool vs alternatives. The descriptions lack the 10 - 1024 character baseline (most are 500+ chars of free-form markdown), making them token-wasteful and unfocused for agent reasoning. No parameter descriptions exist because no parameters are defined. Error handling is minimal (generic try/catch returning formatted text). The server provides no pagination, no output schemas, no structured response fields for chaining. Tools are read-only information getters, low risk, but even informational tools should have documented return structures.
Get all available information topics
Get information about Docker containerization platform
Get information about Docker MCP Catalog
Get information about Docker MCP Toolkit
Get information about Model Context Protocol (MCP)
Get information about Kristiyan Velkov
All 6 tools have empty input schemas (properties: {}). Per HARD SCORING RULES, schema score MUST be 0 when schemas lack type definitions. Tools should document input expectations, even if 'no parameters accepted', use additionalProperties: false explicitly.
Descriptions are 500 - 800 character markdown blobs, well above the 10 - 1024 character baseline. A tool description should answer: WHAT, WHEN, WHAT RETURNS in 50 - 150 characters.
No output schema documented. The response structure is inferred from return statements (content array with text type). Tools should explicitly define: 'Returns an object with {content: [{type: string, text: string}]}' so LLMs know what to expect and can chain calls.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 30 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 0 | - | v1 |
who_is_kristiyan_velkov violates naming guidance. The name does not reflect a clear action verb, 'who_is' is ambiguous. 'get_author_info' or 'describe_kristiyan_velkov' would be clearer and match LLM expectations for action-verb prefixes.
No tool distinguishes when to call what_is_mcp vs what_is_docker_mcp_toolkit vs what_is_docker_mcp_catalog. Descriptions do not explain: 'Call this for X-level detail' or 'Call this when the user asks about Y.' LLMs will guess, risking wrong tool selection.
Error handling is generic: try/catch returns 'Error: {message}' with no recovery guidance. Per pattern:recovery-guide, errors should tell the LLM what to do next. E.g. if a tool fails, offer an alternative.