8 tools with complete JSON schemas and descriptions present. Naming follows verb_noun pattern (search, count, get, index, delete, bulk, msearch, knn). Descriptions are concise (40-120 chars) but lack actionable context for LLM selection, they state WHAT but not WHEN or WHY. Parameter descriptions exist but are minimal (e.g., 'Target index' lacks guidance on allowlist constraints). No output schemas documented. Error handling and recovery guidance absent. Security model (index allowlist, SSRF guard, cred:// auth) is implemented but not surfaced in tool descriptions. Missing idempotent/destructive hints despite having WRITE and DESTRUCTIVE risk classifications.

• Document output schemas for all 8 tools. Example for search: {type: 'object', properties: {hits: {type: 'array', items: {type: 'object', properties: {_id, _source, _score}}}, total: {type: 'integer'}, aggregations: {type: 'object'}}}. This enables LLMs to extract _id for follow-up get/delete calls.
• Expand descriptions to 80-150 chars with WHEN/WHY. Example: 'Search documents using Elasticsearch DSL. Use for complex queries with filters, aggregations, or sorting. For vector similarity, use knn instead. For multiple independent queries, use msearch.'
• Add parameter descriptions explaining constraints and error cases. Example for index: 'Target index name (must be in the binding allowlist; if omitted, defaults to default_index configured in binding spec). Invalid index returns 403 Forbidden with list of allowed indices.'
• Implement tool annotations in schema. Add readOnlyHint: true to search/count/get/msearch/knn; destructiveHint: true to delete/bulk; idempotentHint: true to get/search/count/knn (read-only ops are idempotent).
• Add error recovery guidance. Example for delete: 'If document not found, returns 404. If index not in allowlist, returns 403 with available indices. Retry with correct index or call search() first to verify document exists.'