May-la MCP Server provides three well-scoped, read-only file navigation tools with clear verb-based names and reasonable descriptions. All tools have explicit JSON Schema input definitions with typed parameters and defaults. However, there are notable gaps: (1) output schemas are entirely undocumented, callers cannot see what fields to expect from each tool's response, (2) parameter descriptions lack specificity around constraints and formats (e.g., what glob patterns are supported, what 'tech stack detection' means), (3) error handling is generic with no recovery guidance, (4) no guidance on dependencies between tools or when to call each one. The server lands in the C-to-B range: above median due to proper schemas and clear naming, but held back by missing output documentation and sparse parameter guidance.

• Document the output schema for each tool. For 'find': state that it returns an object with 'results' array containing objects with 'path', 'name', 'type', 'size', 'modified' fields. For 'info': list what gets returned (e.g., 'tech_stack', 'tree', 'entry_counts', 'entry_points'). For 'read': specify returned fields (e.g., 'content', 'summary', 'line_count', 'symbol_location').
• Expand parameter descriptions with concrete constraints. For 'pattern' in find: 'Glob pattern using doublestar syntax (**, *, ?, [abc]). Examples: **/*.go, src/**/*.ts, Dockerfile*'. For 'depth' in info: 'Max tree depth (1-10, default 4), deeper trees may timeout on large projects'.
• Add error recovery guidance in tool descriptions. E.g., 'find' could note: 'If no results found, verify the pattern is valid (test with simpler patterns like *.go). If timeout occurs, reduce max_results or check if the root path is accessible.'
• Clarify 'read' tool parameter precedence. Document: 'If both lines and symbol are provided, symbol takes precedence. Batch mode (comma-separated paths) reads all files and returns results keyed by path.'
• Add discovery hints. For 'find': 'Use this after info() to locate specific files. Combine with read() to inspect them.' For 'info': 'Start here to understand project structure before targeted file searches.'
• Document max_results default and timeout. 'Maximum number of results (default 200). Large result sets may slow response; use pagination by reducing max_results if needed.'
• Specify supported encoding for 'read'. The code shows UTF-8, UTF-16LE, UTF-16BE support, document this: 'Automatically detects and converts UTF-8 (with/without BOM), UTF-16LE, UTF-16BE. Binary files are best-effort decoded as Latin-1.'