This Meta Ads MCP server has 8 tools with mostly complete schemas and adequate descriptions, but falls short of production quality in several critical areas. All tools have input schemas with proper types and descriptions, which is a strong baseline. However, output schemas are not documented, we cannot verify what structure each tool returns, making it impossible for LLMs to chain results confidently. Descriptions are present but generic and under-optimized for LLM decision-making (averaging ~70 chars, below the 194-char baseline for A+ tools). Parameter names are mostly clear (verb-noun style), but some tools expose access_token as a parameter despite the guideline that credentials must never appear as tool parameters, this is a critical security issue. Error handling is not evident in the visible code; there is no recovery guidance or error classification. The create_ad tool lacks important details (no enum for status values, no examples of valid tracking_specs format). Tools are well-composed (each does one thing), but the server lacks pagination documentation for list-returning tools. Overall, this is a functional but baseline-quality implementation.

• Remove 'access_token' from all 8 tool parameters. Implement server-side token injection: read token from environment variable (e.g., META_ACCESS_TOKEN) at startup and pass it to API calls internally. Document in README how to configure the token.
• Document output schema for each tool. Example: get_ad_accounts should specify it returns {"data": [{"id": string, "name": string, "currency": string, ...}], "paging": {"cursors": {...}}}. Make fields match parameter names of dependent tools.
• Add enum constraint to create_ad 'status' parameter: {"enum": ["ACTIVE", "PAUSED", "ARCHIVED", "DELETED"]}. Remove example from description.
• Expand tool descriptions to 150-250 chars each. Add WHEN-to-use guidance. Example: 'Get detailed info about a Meta Ads account (budgets, spend, status). Use this after get_ad_accounts to inspect a specific account before making changes.'
• Add max constraint to 'limit' parameters. Document: 'limit': max 200 for get_ads, max 100 for search_ads_archive. Clarify if results are paginated (cursor or offset-based).
• Replace 'tracking_specs' example in description with formal schema. Define accepted structure as JSON Schema object or link to Meta's official tracking spec documentation.
• Implement error handling with recovery guidance. Return structured errors: {"error": "user_error", "message": "Invalid account_id format. Expected 'act_XXXXXXXXX', got 'account-123'. Valid accounts: [act_001, act_002]"}.
• Add a confirm_create_ad variant or require explicit 'confirm' parameter for create_ad to prevent accidental ad creation.

Inferred effective spec: 2026-07-28+.

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