Bash-based STDIO server with 6 tools. Tool names lack action verbs (meta_get, meta_post, meta_delete are acceptable but generic). Descriptions are present but minimal (10-50 chars, below the 50-200 char LLM-optimized baseline). Input schemas visible for all tools with basic type info, but parameter descriptions are terse and lack constraints. No output schemas documented. Error handling is absent, tools fail silently or return raw API errors without recovery guidance. No security annotations (readOnlyHint/destructiveHint present in metadata but not in schema). Composition is reasonable (separate GET/POST/DELETE), but parameter naming lacks clarity (e.g., 'query' and 'body' as strings rather than structured objects).

• Expand tool descriptions to 100-150 characters. For meta_get: 'Fetch data from Meta Graph API (e.g., ad account insights, campaign performance). Use for read-only queries. Requires valid access token and appropriate permissions (ads_read, pages_read_engagement, etc.).'
• Document output schemas for each tool. Example for meta_get: 'Returns JSON object with data array (if paginated) or direct fields. Includes pagination cursors (paging.cursors.before/after) if applicable. Always includes error field if request fails.'
• Add parameter constraints. For 'path': 'Graph API endpoint path (e.g., me, me/adaccounts, act_123456/insights). Must start with a valid node ID or alias. No leading slash.'
• Implement error classification and recovery guidance. Catch Meta API errors and return: '{"error": true, "code": 190, "message": "Token expired", "recovery": "Regenerate system-user token in Business Settings → System users. See SETUP.md.", "retryable": false}'
• Add tool annotations to schema. Mark meta_post and meta_delete with destructiveHint: true; mark meta_get and meta_page_get with readOnlyHint: true. This signals to LLMs which tools have side effects.