Server has 5 tools with mixed quality. Three tools (web_fetch, web_search, listdir) have decent parameter schemas but weak descriptions that lack context for LLM decision-making. graphmert_write and graphmert_search have minimal descriptions. None of the tools document their output schemas explicitly. Parameter descriptions are present but inconsistent, some are specific (e.g., 'Maximum bytes to retrieve'), others generic. Tool naming is acceptable (verb-noun convention), but descriptions fail to explain WHEN to use each tool vs. alternatives, or what the LLM should expect in responses. Error handling is not visible in the tool definitions. No tool annotations (readOnlyHint/destructiveHint) despite clear intent (graphmert_write is a mutation). This is a typical community server in the C-range: functional but lacking LLM-optimization and production polish.

• Document the complete output schema for each tool. For web_fetch, specify: {url: string, text: string, links: string[], metadata: {title?: string, encoding?: string, size: integer}}. Use JSON Schema format in comments or a separate schema definition.
• Expand tool descriptions to 150 - 200 chars following the WHEN/WHAT/WHY pattern. Example for web_fetch: 'Fetch a URL and extract plain text. Use when you need current content from a webpage (e.g., documentation, news). Strips HTML and enforces a size limit. Returns text and optional link list. Timeouts after 15s by default.'
• Add tool annotations: mark graphmert_write with destructiveHint: true, graphmert_search with readOnlyHint: true. This guides agents on safety and idempotency.
• Replace pipe-delimited enums in descriptions with formal JSON Schema enums. For listdir, define: 'sort_by': {type: 'string', enum: ['name', 'size', 'modified']}. For web_search, define: 'engine': {type: 'string', enum: ['auto', 'bing', 'google', 'brave', 'serpapi', 'searxng']}.
• Clarify parameter formats. For web_search 'lang', document: 'ISO 639-1 language code (e.g., en, fr, de)'. For graphmert_write 'who', clarify: 'Username or identifier of the memory author (string, no spaces recommended)'.
• Add error handling guidance to descriptions. For web_fetch: 'If timeout occurs, increase timeout_s or try breaking the request into smaller chunks. On 4xx errors, verify the URL is correct and public.'
• For graphmert_write, document the response: does it return a memory_id? metadata? Do you need the ID for subsequent operations? Current description leaves this unclear.

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

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