Webfang provides 16 tools with consistent naming (verb_noun pattern) and documented schemas. However, descriptions are often generic (50-100 chars), parameter descriptions lack detail about constraints/formats, and output schemas are not explicitly documented. Error handling guidance is minimal. Most tools follow basic patterns but lack the depth expected for production-grade agent tooling. No tool annotations (readOnlyHint/destructiveHint) despite clear risk classifications in metadata.

• Document output schemas for all tools. E.g., scrape_url should declare: returns {content: string, title?: string, links?: string[], metadata?: object}. This enables LLM planning and tool chaining.
• Convert parameter constraints to JSON Schema enums and patterns. E.g., format parameter in scrape_url: enum=['markdown', 'html', 'text', 'json']. Limit parameter: minimum=1, maximum=100. Remove examples from descriptions.
• Expand tool descriptions to 150-250 chars. Include: what it does, when to use it vs similar tools, and what structure it returns. E.g., 'Scrape content from a URL using CSS selectors. Returns structured data (title, content, links) in markdown, HTML, text, or JSON. Use for content extraction; pair with extract_links for link discovery.'
• Add tool annotations via rmcp macros. Mark open_in_obsidian with #[destructive], read-only tools with #[read_only], and idempotent tools with #[idempotent]. This enables agents to reason about retry safety and side effects.
• Add error recovery hints to descriptions. E.g., detect_obsidian_vault: 'If vault not found, check vault_name spelling or call list_vaults() to discover available vaults.' This guides agent self-correction.

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

• Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained