A Model Context Protocol (MCP) server providing access to the Australian Pharmaceutical Benefits Scheme (PBS) API for retrieving information about medicines, pricing, and availability.
The PBS MCP server defines a single tool 'pbs_api' with a moderately detailed schema and description. The tool name follows verb_noun conventions ('pbs_api' is somewhat generic but acceptable for API gateway patterns). The description is adequate at ~140 chars but lacks guidance on WHEN to use this tool vs alternatives, and does not clarify prerequisites or error recovery paths. The input schema is well-structured with proper JSON Schema types, enums for HTTP methods, and descriptions for most parameters. However, parameter descriptions are generic and lack concrete format/constraint details. The tool lacks error classification guidance, output schema documentation, and pagination information. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) are present. Output structures are not documented, making it harder for LLMs to plan downstream steps.
Access the Australian Pharmaceutical Benefits Scheme (PBS) API to retrieve information about medicines, pricing, and availability.
Tool name 'pbs_api' is generic and does not follow verb_noun convention. Should be 'query_pbs_api', 'get_pbs_data', or similar to clarify the action. Current name does not convey to LLM what operation occurs when invoked.
Tool description lacks WHEN guidance. It states WHAT (access PBS API) but not WHEN to call this tool, what prerequisites exist, or how to recover from failures. Missing actionable recovery hints (e.g., 'If endpoint returns 401, verify subscriptionKey is valid').
Output schema is not documented in code or descriptions. LLMs cannot know what structure pbs_api returns, forcing them to guess at response fields and plan downstream steps blindly. Must document return type schema.
Parameter 'params' uses generic 'additionalProperties: {type: string}' without validation rules, format constraints, or examples. LLMs cannot determine valid query parameter names or expected value formats. Should document PBS API-specific parameter constraints or link to PBS API documentation.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 53 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 38 | - | v1 |
Parameter descriptions are minimal. 'endpoint' lacks examples of valid endpoint names beyond the brief comment. 'params' gives one example but no format or enumeration. 'subscriptionKey' description does not explain how default public key is obtained or when custom keys are needed.
No error classification or recovery guidance. Tool can fail (network timeout, invalid endpoint, rate limit, auth failure) but error messages do not guide LLM on retry strategy, fallback tools, or user-facing explanations.
No tool annotations present. The tool is read-only (does not modify PBS data), but MCP readOnlyHint annotation is missing. This leaves the LLM uncertain whether calling this tool has side effects.
subscriptionKey parameter exposed as a tool input parameter. This violates secret-injection pattern, credentials must not appear in tool parameters where they can be logged in traces. Should use server-side environment variable injection instead.
No pagination support documented. If PBS API returns large result sets, tool lacks limit/offset or cursor parameters. LLM may request unbounded result sets, exhausting context or hitting API limits.