Loading ToolBench

Reading the current corpus…

athlete-performance-platform-showcase MCP server · F · ToolBench

athlete-performance-platform-showcase

github.com/ellttdave/athlete-performance-platform-showcase

Demonstrates LLM tool calling with Anthropic Claude API, MCP-like router pattern for tool execution, and RAG (Retrieval-Augmented Generation) implementation with vector embeddings

2026-07-28+STREAMABLE_HTTP2 tools

36/100

Definition Quality45

Protocol Readiness15

Supportability36

Spec posture

Inferred effective spec 2026-07-28+·current baseline 2026-07-28

Static source inference · medium confidence · evidence: Streamable HTTP

Current-spec patterns detected

Summary

This is a showcase/demo implementation, not a production MCP server. Tools are defined in custom TypeScript interfaces (ToolDefinition) rather than via MCP protocol. The two tools have basic descriptions but lack critical schema rigor: no explicit input validation, no output schema documentation, no error recovery guidance. Parameter descriptions are minimal (entity_id, time_period_days lack format/range constraints). No tool annotations (readOnlyHint, destructiveHint). The codebase shows LLM integration patterns but does not implement MCP protocol compliance. Tool definitions are inferred from code rather than formally registered via MCP.

Tools (2)

analyze_dataread onlysource verified50/100

Comprehensive data analysis tool with multi-modal data integration

get_summaryread only40/100

Get quick summary of entity data

Issues (5)

CRITICALtool ↗

No output schema documented. Tools return unstructured JSON without field definitions, forcing LLMs to parse results blindly. analyze_data returns 'aggregatedData' (undefined structure); get_summary returns '{summary: string}' with no type safety.

analyze_dataget_summary

HIGHconstrained-input ↗

Parameter descriptions lack constraints. 'time_period_days' has no min/max bounds (default 730 suggests 0-730 range, but not stated). 'entity_id' has no format guidance (UUID? alphanumeric? length limits?). LLMs will pass invalid values without explicit constraints.

analyze_data

HIGHrecovery-guide ↗

No error recovery guidance. Router returns generic errors ('Tool execution failed', 'Unknown tool') without actionable next steps. LLMs cannot distinguish retryable vs fatal failures or know what to try next.

analyze_dataget_summary

CRITICALtool ↗

Recommendations

Document output schemas explicitly. For analyze_data, define: {aggregatedData: {metrics: object, trends: array, anomalies: array, timestamp: string}}. For get_summary, define: {summary: string, entity_id: string, generated_at: string}.
Add parameter constraints to descriptions. Change 'time_period_days' description to: 'Number of days to look back (1-730, default: 730). Must be a positive integer.'
Implement error recovery guidance. When entity_id is not found, return: 'Entity "xyz" not found. Call list_entities() to discover available IDs.' When time_period_days exceeds 730, return: 'Maximum lookback is 730 days. Reduce time_period_days or call get_summary() for a quick overview.'
Add tool annotations to router. Modify TOOL_REGISTRY entries to include: {description, handler, readOnlyHint: true, idempotentHint: true} for both tools (both are read-only and safe to retry).
Migrate to official MCP protocol. Replace custom HTTP router with MCP SDK (Node.js). Implement proper tool registration via server.setRequestHandler(ListToolsRequestSchema, ...) and CallToolRequestSchema handlers. Add _meta with protocol version and capabilities in each response.

Score history

Overall score trend

First recorded score · v2 rubric

36/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	F	36	2026-07-28+	v2