kbolt provides 9 read-only tools with complete input schemas and descriptions. Naming follows verb_noun convention (search, get, list, etc.). All tools have descriptions (avg ~60 chars) and documented parameters with types. However, descriptions are terse and lack LLM-optimized guidance on WHEN to use each tool vs alternatives. Output schemas are not documented, LLMs cannot predict response structure. No error handling guidance. Parameters like 'identifier' in get() accept multiple formats (path, docid, chunk locator) but lack explicit format documentation. The multiget tool's 'locators' parameter description says 'comma-separated' but the schema shows array type, inconsistency. No tool annotations (readOnlyHint, etc.) despite all being read-only. Overall: functional but minimal, typical of a CLI-first tool adapted to MCP without LLM-specific optimization.

• Document output schemas for all tools. Include field names, types, and examples. E.g., search() returns {results: [{docid, path, score, content}], total_count, elapsed_ms}.
• Expand tool descriptions to 100-150 chars with LLM-optimized guidance: WHAT it does, WHEN to use it vs similar tools, and any prerequisites. E.g., 'search: Find documents by keyword or semantic similarity. Use for discovery; use get() if you already know the docid.'
• Fix multiget 'locators' parameter: clarify whether it accepts a comma-separated string or JSON array. Update description to match schema type.
• Add readOnlyHint: true to all tool definitions (via tool annotations in MCP 2026-07-28 spec) to signal idempotency and safe retry semantics.
• Add error recovery hints to tool descriptions. E.g., space_info: 'If space not found, call list_spaces() to see available spaces.' This guides agents on self-correction.