Webpipe has 23 tools with visible schemas and descriptions, but quality is inconsistent. Core tools (search_evidence, web_fetch, web_extract) have detailed parameter schemas with 10-20 params each, but descriptions are terse (50-100 chars) and lack actionable guidance. 6 deprecated tools (web_search, arxiv_search, http_fetch, etc.) clutter the surface. Parameter descriptions are generic ('Search query string', 'URL to fetch') without format constraints, ranges, or examples of valid values. No output schemas documented. Error handling is absent, no guidance on retryability, user-fixable errors, or recovery paths. Tool naming is reasonable (verb_noun pattern) but composition is poor: search_evidence, web_fetch, web_extract, and web_seed_search_extract overlap significantly, forcing LLMs to reason about which to use. No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels. Baseline: 549 production tools average 194 chars for descriptions; webpipe averages ~70 chars.

• Remove or hide deprecated tools (web_search, arxiv_search, http_fetch, page_extract*). If backward compatibility is required, mark them as 'deprecated' in descriptions and redirect to canonical tools.
• Expand parameter descriptions to 50-150 chars each. Include format constraints, ranges, and defaults. E.g., 'Search query (1-500 chars, alphanumeric + spaces). Default: empty string.' instead of 'Search query string'.
• Document output schemas for all tools. Use JSON Schema or a structured format showing field names, types, and meanings. E.g., 'Returns {results: [{url, title, snippet, relevance_score}], total_count, next_cursor}'.
• Add error handling guidance to tool descriptions. E.g., 'Timeouts are retryable. Invalid URLs return a 400 with the constraint violated. Network errors are transient, retry with exponential backoff.'
• Add tool annotations (readOnlyHint, destructiveHint, idempotentHint) to schemas. Mark search_evidence, web_fetch, web_extract as readOnly. Mark repo_ingest, webpipe_usage_reset as destructive.