Loading ToolBench
Module pour l'intégration MCP Payplug. A PHP library for integrating Payplug payment processing with MCP (Model Context Protocol).
This PHP library exposes 5 tools for Payplug payment operations. Critical gaps: (1) Two tools (abortAction, captureAction) are empty placeholders with zero implementation and no input schemas, they cannot function. (2) Three implemented tools have descriptions but lack formal input schema definitions visible in the source; schemas are inferred from parameter documentation only. (3) No output schemas documented. (4) Tool names use verb_noun pattern (good), but descriptions are generic and lack LLM-optimized guidance on when to use each tool vs. alternatives. (5) Parameters like 'api_bearer' expose secrets as tool inputs, violates secret-injection pattern. (6) No error handling guidance, recovery paths, or actionable error messages. (7) Composition is reasonable (separate create/refund/retrieve), but no pagination, batch operations, or idempotency guarantees documented.
Placeholder method for aborting a payment. Currently not implemented.
Placeholder method for capturing a payment. Currently not implemented.
Creates a payment resource using the Payplug API. Accepts a PaymentInputDTO with payment method, amount, currency, customer details, URLs, and optional metadata/context. Returns a PaymentOutputDTO with the created payment resource or error details.
Creates a refund for a paid payment resource. Validates the refund input DTO and payment resource, then processes the refund through the Payplug API. Returns a RefundOutputDTO with the refund resource or error details.
Retrieves an existing payment resource from the Payplug API by its resource ID. Returns a PaymentOutputDTO with the payment resource details or error information.
Two tools (abortAction, captureAction) are unimplemented placeholders with empty bodies, no descriptions, and no input schemas. They cannot be invoked and should be removed or completed.
API bearer token exposed as a required tool parameter (api_bearer). Secrets must never appear in tool parameters, use server-side secret injection via environment variables or vault. Agent traces log all parameters, leaking credentials.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., payment_id for refund) without knowing response structure.
Input schemas are inferred from parameter descriptions only; no formal JSON Schema definitions visible in source code. Schema validation and LLM parameter selection rely on unstructured text.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 24 | 2026-07-28+ | v2 |
No error handling guidance. Tools lack recovery hints, error categorization (retryable vs. fatal), or actionable messages. An LLM receiving a generic exception has no path forward.