SEO for Agents — Score any AI agent endpoint on trust & capability metrics
Three tools with complete schemas and descriptions, but significant gaps in parameter documentation and output schema clarity. Tool names follow verb_noun convention (score_agent, compare_agents, get_fix_recommendations). Descriptions are detailed (150-250 chars) and explain WHEN to use each tool. However, parameter descriptions are minimal or missing context. Output schemas are not formally documented, the code shows text formatting but no structured schema declaration. Error handling is implicit (relies on scan_agent_v2 internals). No input validation rules stated in descriptions (e.g., URL format constraints, timeout behavior). Composition is sound: three distinct, single-purpose tools that chain naturally (score → compare → fix recommendations).
Choosing between two AI agents for a task? Scores both and shows which is stronger in each category — schema quality, reliability, docs, ecosystem, maintenance. Helps pick the more trustworthy option based on data, not guesswork.
Building an AI agent? Get a prioritized list of improvements to make it more discoverable and trustworthy. Shows current score, what to fix, expected point gains per fix, code templates, and spec links. Use this before releasing your agent to maximize its discoverability.
Before integrating an AI agent, check if it's trustworthy. Before releasing your own agent, check if it's discoverable. Scores any MCP server or AI agent URL on 5 dimensions: Schema Quality, Functional Reliability, Developer Experience, Ecosystem Signal, and Maintenance Health. Returns 0-100 score with grade (A-F), category breakdown, and prioritized fix recommendations showing exactly what to add for maximum improvement.
Output schemas not formally documented. Code shows text formatting but no structured schema declaration for tool responses. LLMs cannot plan downstream calls or extract typed fields.
Parameter 'url' lacks format constraints. No mention of HTTPS requirement, URL validation, timeout behavior, or what happens if the endpoint is unreachable. Description says 'HTTPS URL' but no validation rules stated.
No error recovery guidance. If scan_agent_v2 fails (network error, invalid URL, timeout), the tool returns an error but does not tell the LLM what to do next or how to retry.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | 2026-07-28+ | v2 |
'skip_mcp' parameter description is terse ('Skip MCP protocol handshake'). Does not explain when to use it, performance tradeoff, or what 'HTTP checks only' means. LLM cannot reason about when to set it.
Output structure (categories, checks, top_fixes, mcp metadata) is hardcoded in formatting logic but not declared as a schema. If the LLM needs to extract 'grade' or 'total_score' programmatically, it must parse text.