Server has 5 tools with mixed quality. All tools have descriptions and basic schemas, but critical gaps exist: (1) Output schemas are undocumented for 4 of 5 tools, LLMs cannot plan downstream calls or extract results reliably. (2) Parameter descriptions are minimal (1-2 words) and lack format/constraint guidance. (3) Error handling is generic ('error' field) with no recovery guidance. (4) GCP tools lack permission/scope declarations. (5) The 'add' tool mixes Boomi integration concern with arithmetic, violating single-responsibility. Baseline: 194 chars avg description; these average ~40 chars. Baseline: 100% A+ tools document return types; this server documents 1/5.

• Add explicit output schemas to all GCP tools in gcp_mcp_server.py. Use FastMCP's return type hints or add a 'returns' field to each @mcp.tool() decorator documenting the dict structure (e.g., {'instances': [{'id': str, 'name': str, ...}], 'count': int, 'status': str}).
• Expand parameter descriptions to 50-100 chars each. Example: 'project' → 'GCP project ID (e.g., my-project-123). Find yours in the GCP Console under Settings > Project ID.' This matches the 72-char baseline and provides actionable guidance.
• Add error recovery hints. Replace generic error returns with: {'error': 'Invalid project ID: my-project-xyz. Call list_projects() to see available projects.', 'status': 'error', 'retryable': False}. This guides LLM recovery per pattern:recovery-guide.
• Declare required IAM permissions for each GCP tool. Add a 'permissions' field to tool metadata: 'permissions': ['compute.instances.list'] for list_instances, etc. This enables least-privilege agent configuration.
• Clarify the 'add' tool's relationship to Boomi. If it's a Boomi-specific integration, rename to 'add_via_boomi' and document Boomi prerequisites. If it's just arithmetic, remove the Boomi reference and simplify the description to 'Add two integers and return the sum.'