Two tools are registered with FastMCP and have descriptions visible in source code. However, both tools exhibit significant quality gaps: descriptions lack critical context about when to use them vs. alternatives; input parameters have minimal documentation; output formats are poorly specified; no error handling guidance is provided; and no defensive validation is shown. The naming follows verb_noun convention (covid_faq_retrieval_tool, firecrawl_web_search_tool), which is acceptable, but the descriptions are vague about resource selection logic and prerequisites. Parameter schemas are minimal (only 'query' string with basic description). No output schema is documented, the first tool returns 'str' and the second returns 'List[str]', but the actual structure of returned content is opaque to an LLM. Error handling is present in the code (type checks, logging) but does not guide the LLM on recovery ('No embeddings matched' returns a fallback message, but there is no structured error response). This is a below-average server typical of early-stage RAG tooling.

• Document the output schema for each tool. For covid_faq_retrieval_tool, specify that the return is 'A string containing concatenated context snippets from matched documents, separated by spaces.' For firecrawl_web_search_tool, specify the structure of each list element (e.g., 'A list of URLs retrieved from web search results, or a list of search result summaries'). If possible, return structured objects with fields like {url, title, snippet, relevance_score} instead of raw strings.
• Clarify tool selection logic. Add explicit selection criteria: 'Use covid_faq_retrieval_tool for questions about COVID-19 facts, vaccines, symptoms, transmission, or safety guidelines. Use firecrawl_web_search_tool for questions about current events, non-COVID topics, or real-time information.' Include an example: 'User asks "Is COVID still contagious?" → covid_faq_retrieval_tool. User asks "What are the latest COVID variants in 2025?" → firecrawl_web_search_tool.'
• Expand parameter descriptions to include constraints and formats. Example: 'query (string, 1 - 500 characters, required): The user question or search phrase. Use natural language. Avoid special characters unless they are part of a medical term (e.g., "COVID-19").'
• Add error handling guidance. Specify what errors are possible and what the LLM should do: 'If the tool returns "I couldn't find a relevant answer", try rephrasing the question or use firecrawl_web_search_tool for broader coverage. If a connection error occurs, wait 5 seconds and retry.'
• Add result limits and pagination. For firecrawl_web_search_tool, specify: 'Returns up to 10 most relevant results. To get more, call the tool again with a refined query or pass a page parameter (if supported).' Document any pagination parameters in the schema.

Inferred effective spec: <=2025-11-25.

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