Project Tendril provides 7 browser automation and web research tools with complete input schemas and descriptions. All tools have verb-noun naming (navigate, snapshot, click, fill, search, crawl, fetchText) and clear parameter documentation. However, descriptions are terse (10-50 chars), lacking context on WHEN to use each tool or dependencies between them. No output schemas are documented, LLMs cannot predict what fields to expect from responses. Error handling is not visible in the tool definitions. The crawl tool lacks pagination guidance despite potentially returning large result sets. No tool annotations (readOnlyHint/destructiveHint) are present despite clear risk classifications in the metadata.

• Expand tool descriptions to 50 - 200 characters. Example: 'navigate' → 'Navigate to a URL and wait for page load. Use this to start a browsing session or move to a new page. Specify waitUntil to control when navigation is considered complete (domcontentloaded for fast feedback, networkidle for full page load).'
• Document output schemas for all tools. Example: 'snapshot returns {html: string, mode: string, timestamp: ISO8601}' or 'search returns {results: [{title, url, snippet}], total: number, nextCursor?: string}'.
• Add tool annotations to the MCP server registration. Mark click and fill with destructiveHint=true; mark navigate, snapshot, search, crawl, fetchText with readOnlyHint=true.
• Add pagination guidance to crawl. Document that results are streamed or paginated, and provide a cursor/offset mechanism if result sets exceed 50 items.
• Document error scenarios and recovery paths in descriptions. Example: 'If navigation times out, the page may be slow or unreachable. Retry with a longer timeout or check the URL.'

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

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