The Glean MCP server defines 3 tools with reasonable naming conventions and basic input schemas. All tools are read-only and have descriptions. However, descriptions are minimal (under 100 chars for most), parameter descriptions lack depth, and output schemas are not documented. Tool names follow verb_noun convention (glean_search, glean_chat, glean_read_documents), which is positive. The schema quality is moderate, tools have input types defined but lack detail on constraints, error conditions, and response structures. No complex composition or pagination patterns visible.

• Document complete output schemas for all three tools. Specify field names, types, and which fields are always present. Example: glean_search returns {results: [{id, title, snippet, url, metadata}], total_count, has_next_page}.
• Expand parameter descriptions with actionable constraints. For glean_search: 'Search query string (1-500 chars, natural language). Glean supports boolean operators (AND, OR, NOT) and field filters (owner:name, status:archived).'
• Add bounds to page_size: 'Number of results per page (1-100, default 10). Larger pages reduce API calls but increase latency.'
• Clarify glean_read_documents specs parameter: 'Array of document identifiers. Each spec must contain either id (Glean internal ID) OR url (Glean document URL), not both. Pass id for fastest retrieval if available.'
• Add error recovery hints to each tool. Example: 'Returns empty results if no matches found, try broadening query or removing filters. If authentication fails, verify GLEAN_API_TOKEN is set and not expired.'
• Document pagination for glean_search: 'Response includes total_count (total matching documents) and has_more (boolean). Use offset or cursor-based pagination for large result sets. Max 100 results per page.'
• Expand tool descriptions to clarify use cases: 'glean_search: keyword/filter-based search, returns snippets and metadata. Best for discovery. glean_chat: natural language Q&A, returns conversational responses. Best for complex questions needing reasoning.'
• Validate page_size input and return clear errors: 'page_size must be 1-100, got 150. Reduced to 100.' Let LLM know the correction was made.