Loading ToolBench
Model Context Protocol server for Lyra Intel codebase analysis
The Lyra Intel MCP server has significant gaps in definition quality. Tool schemas are not visible in the provided source code, only tool names and generic descriptions are shown in mcp-server/src/tools/registry.ts. Without seeing the actual parameter definitions, input schemas, and return type documentation, schema quality cannot be assessed. Descriptions are present but generic and lack LLM-optimization guidance. No evidence of parameter descriptions, type constraints, enums, or error recovery patterns. The tools appear functional (read-only code analysis), but the definitions lack the rigor required for reliable agent integration.
Analyze this codebase to understand its structure, identify complexity hotspots, dependencies, and potential security issues.
Check code complexity metrics to identify functions that might be too complex or hard to maintain.
Scan for security vulnerabilities, hardcoded secrets, and compliance issues in the codebase.
Search the codebase for code patterns, functions, or implementations matching a specific query.
No input schemas visible in source code. For all 4 tools, the parameter definitions, types, descriptions, and constraints are not shown.
Descriptions are generic and under-optimized for LLM selection. 'Analyze this codebase to understand its structure...' (59 chars) lacks specificity on WHEN to use vs alternatives, prerequisites, or expected output structure. Baseline A+ tools: 50-200 chars with clear intent discrimination.
No parameter descriptions visible. The source listing shows tool names only; no evidence of per-parameter documentation explaining types, constraints (enum values, ranges, formats), or defaults. Baseline A+ tools: 100% of params have descriptions.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 33 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 0 | - | v1 |
No documented output schemas. Without field-level documentation of return types, the LLM cannot plan downstream tool calls or extract relevant data for multi-turn reasoning.
No error handling guidance visible. If a codebase cannot be analyzed (no Git history, unsupported languages, permission denied), there is no documented recovery path or actionable error message format.
Tool 'analyze-codebase' may conflate multiple responsibilities (structure + complexity + dependencies + security). Best practice: split into separate, single-responsibility tools so the agent can compose them intentionally.
No pagination or result-limiting guidance. Code analysis tools often return large datasets (all functions, all findings). Without limit/offset parameters and documented caps, responses can exceed context window.