DocGraph provides 10 tools with complete JSON schemas and descriptions. Naming follows verb_noun convention (index_project, search, get_document). Descriptions are present but brief (avg 45 chars, below the 194-char baseline). All tools have input schemas with type definitions. However, parameter descriptions are minimal or missing context about when to use each tool. Output schemas are not documented, the code shows `toLeanResult()` compacting SearchResult, but this transformation is not declared in the tool definition. Error handling is absent from tool definitions. The server supports multi-project queries via projectPath parameter, which is well-designed for composition, but lacks guidance on error recovery.

• Expand tool descriptions to 100 - 200 chars. Example: 'search' → 'Search indexed documents using hybrid full-text + vector search. Use this to find relevant docs by topic or keyword. Returns up to 20 results with snippets; call get_document to fetch full text.'
• Document output schemas for each tool. Example for search: 'Returns array of {id, path, title, language, extension, score, snippets[{field, line, text}]}. Use id or path in get_document to fetch full content.'
• Add error recovery hints to descriptions. Example: 'If search returns no results, try explore() for broader context, or call index_project() to ensure the project is indexed.'
• Clarify parameter relationships. For projectPath: 'Optional. Defaults to the server's project root. Pass a different path to query another project in the same session without restarting the server.'
• Declare pagination behavior. For search and list_documents: 'Results are limited to the specified limit (default 20/100). If more results exist, call again with offset or implement client-side pagination via limit parameter.'

Inferred effective spec: 2025-06-18+.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry