Locus is a Swift CLI tool for iOS test mapping, not a production MCP server. All 5 tools have descriptions (10-80 chars, below the 50-200 char LLM-optimized baseline), but input schemas are either missing or severely incomplete. where_is, what_renders, and match_snapshot have minimal parameter descriptions. affected_tests and missing_identifiers lack proper schema documentation. No output schemas are documented. Error handling is absent, no recovery guidance, no categorization. Tool names use underscores (where_is, what_renders) which is acceptable, but descriptions are too terse to guide LLM selection. No evidence of idempotency, pagination, or composition patterns. The codebase shows strong internal engineering (reverse indexing, orphan detection, deterministic serialization) but zero MCP-specific patterns.

• Expand each tool description to 100-200 characters. Include WHAT it does, WHEN to use it (vs similar tools), and WHAT it returns. E.g., 'where_is: Find all UI elements with a given accessibility identifier and list the test files that reference them. Use this to trace element usage across tests. Returns elements array and tests array with file/line locations.'
• Add explicit type, format, and constraint descriptions to all parameters. E.g., affected_tests.ref: 'Git reference to diff against (e.g., main, HEAD~1, or a commit SHA). Defaults to working tree if omitted.' affected_tests.files: 'Array of file paths to treat as changed (overrides git diff). Each path is relative to the source root.'
• Document output schemas for each tool. E.g., where_is returns {elements: [{identifier: string, file: string, line: number}], tests: [{identifier: string, file: string, line: number, testName: string}]}. Include field types and what each field means.
• Add error handling and recovery guidance. E.g., 'If git reference is invalid, try affected_tests with explicit files parameter instead. If accessibility tree dump is malformed JSON, validate it with a JSON parser first.'
• Create a composition guide or tool-chain diagram showing which tools feed into each other. E.g., where_is → match_snapshot (identifier → dump matching), or affected_tests → where_is (changed files → affected identifiers).