MCP server for optimizing ad buyer budgets based on ROI metrics
This is a minimal, single-tool implementation with critical structural deficiencies. The tool 'optimize_budget' has a valid input schema with typed parameters, but lacks essential descriptions, error handling, and output documentation. The entire project appears to be a proof-of-concept rather than a production-grade MCP server, there is no visible MCP server scaffolding, no transport layer, no protocol compliance, and no tool registration mechanism. The tool logic is hardcoded decision trees with no validation or recovery guidance.
Optimize ad budget based on current ROI
No MCP server implementation detected. Source shows only a Python class in client.py with no MCP protocol scaffolding, server registration, or tool definition via MCP SDK.
Tool description is minimal (35 chars: 'Optimize ad budget based on current ROI'). Descriptions under 50 chars typically score 0-40.
No output schema documented. The tool returns a dict with 'recommended_budget' and 'status' fields, but there is no JSON Schema definition describing the return structure. LLMs cannot plan downstream calls without knowing the response shape.
Parameter descriptions are present but extremely sparse. 'current_roi' and 'daily_budget' have 1-line descriptions with no context on expected range, units, or constraints.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 36 | 2026-07-28+ | v2 |
No error handling or validation. The tool performs magic comparisons (current_roi > 2.5, current_roi < 1.5) with no explanation of what these thresholds mean or what happens if invalid data is passed. No guidance for LLM on what to do if result is unexpected.
Hardcoded decision logic without configurability. ROI thresholds (2.5, 1.5) and budget multipliers (1.25, 0.8) are embedded in code with no way for users to customize. This violates the composition pattern, the tool is inflexible and cannot adapt to different business rules.
No idempotence guarantee. If called multiple times with the same inputs, the tool should return the same recommendation. The code does this, but there is no explicit statement in the tool description or contract that this is safe to retry.
No permission or scope declaration. The tool modifies budget recommendations (WRITE risk category), but there is no mention of required permissions, audit logging, or who can call this tool. Missing scope declaration pattern.