A free, unlimited, stable web-search MCP tool for coding agents
Single tool 'web_search' has a well-structured schema with all 5 parameters typed and described. Description is clear and actionable (145 chars), explaining the keyword-query guidance with concrete examples. Schema includes proper constraints (required fields, additionalProperties: false). However, output schema is not documented, callers cannot see what fields to expect in results. Error handling returns JSON with an 'error' field but lacks recovery guidance. No tool annotations (readOnlyHint, idempotentHint) despite being a read-only, idempotent operation.
Search the web and return results (title, URL, page summary, site name, favicon). Form search_query as concise keywords, not a full question — keyword queries return more relevant results.
Output schema not documented. Callers cannot see what fields (title, URL, summary, favicon, etc.) the tool returns, forcing LLMs to guess structure and plan downstream operations blindly.
Error responses lack recovery guidance. When search fails, the tool returns {"error": "..."}. LLMs need actionable next steps: 'Try a simpler query', 'Check domain filter syntax', etc.
Tool annotations missing. web_search is read-only and idempotent but does not declare readOnlyHint or idempotentHint. Agents cannot infer safety properties without explicit hints.
Parameter 'search_recency_filter' and 'location' lack enum constraints. Descriptions mention valid values (oneDay|oneWeek|oneMonth|oneYear|noLimit; cn|us) but no formal enum in schema. LLMs may hallucinate invalid values.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 62 | <=2025-11-25 | v2 |
Parameter 'content_size' lacks enum constraint. Description mentions 'medium' or 'high' but no enum in schema. LLMs may pass invalid values like 'large' or 'small'.