An MCP (Model Context Protocol) server that exposes research-related tools for web search, fact retrieval, and claim validation. Designed to be consumed by any MCP-compatible client or LLM application.
Three tools with basic descriptions and minimal parameter documentation. All tools have input schemas with type definitions (query, topic, claim as strings), but lack parameter descriptions entirely. Tool descriptions are present (194-250 chars) but generic and lack actionable guidance on when/why to use each tool. No output schemas documented. Error handling is absent, tools return JSON strings without guidance on failure modes or recovery. No tool annotations (readOnlyHint, etc.). Simulated implementations limit production readiness.
Get curated facts about a specific topic. Returns verified factual information useful for research. In a real implementation, this would query a knowledge base or facts API with citations.
Validate whether a claim seems reasonable based on common patterns. Returns a validation assessment with confidence level. This is a simplified validation that checks for presence of source citations (good indicator), absolute language (potential red flag), and claim structure and length. In a real implementation, this would cross-reference with fact-checking databases and use ML models.
Search the web for information about a query. Returns relevant search results with titles, snippets, and sources. In a real implementation, this would connect to a search API like Google, Bing, or DuckDuckGo.
Parameter descriptions missing entirely. 'query', 'topic', and 'claim' parameters have no descriptions explaining format, constraints, or examples. LLMs cannot infer parameter semantics from names alone.
Output schemas not documented. Tools return JSON strings but no schema is provided to LLMs describing the structure (fields, types, required keys). Agents must parse unstructured output, increasing errors.
No error handling or recovery guidance. Tools return JSON but provide no error classification (retryable vs fatal) or actionable next steps. Agents cannot self-correct on failures.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 61 | 2026-07-28+ | v2 |
Tool descriptions lack actionable context. Descriptions state WHAT tools do but not WHEN to use them or how they differ from each other. 'Search the web' vs 'Get facts' distinction is unclear to LLMs.
No tool annotations. Tools lack readOnlyHint, destructiveHint, or idempotentHint metadata. LLMs cannot determine safety properties or retry behavior.