Single tool server with minimal definition quality. Tool name is clear and action-oriented (search_confluence), but description lacks depth on use cases and constraints. Input schema is present with basic type information, but output is returned as plain string without structured schema documentation. Error handling exists but lacks actionable guidance for LLMs. No parameter validation constraints documented.

• Restructure tool output from plain string to JSON object: {results: [{id, title, type, space, url}], count, limit}. This enables downstream tool chaining and structured LLM reasoning.
• Enhance tool description: 'Search Confluence pages using Confluence Query Language (CQL). Useful when user provides a search term and needs to find relevant documentation, wiki pages, or knowledge base articles. Returns up to 10 results with title, space, and URL. Note: Query should be a simple search term or valid CQL syntax, use search_confluence before creating or linking pages.' This provides use-case context and prerequisites.
• Add parameter constraints to query: 'The search term or CQL query (1-500 characters). Use simple keywords or valid CQL syntax (e.g., text ~ "keyword" AND space = MYSPACE). Wrap multi-word queries in quotes.'
• Improve error messages: Instead of generic 'Confluence client not initialized', return: 'Confluence credentials not configured. Ensure CONFLUENCE_URL, CONFLUENCE_USERNAME, and CONFLUENCE_API_TOKEN environment variables are set. Check server startup logs for configuration errors.'
• Add response documentation in docstring: 'Returns structured object with fields: results (array of {id, title, type, space, url}), count (number of results found), query_summary (the interpreted search query). Limit is 10 results per request.'
• Add validation for empty query: Return actionable error: 'Query cannot be empty. Provide a search term (e.g., "API design") or CQL expression (e.g., text ~ "bug" AND type = page).'