A recent analysis of 36 popular Model Context Protocol (MCP) servers revealed that a third of them are failing to be usable by AI agents, despite passing compliance checks. The study, which used a new tool called mcpgrade, found that issues were not with the protocol itself but with the quality of tool descriptions, naming, and schema design. A significant problem identified was an epidemic of undocumented parameters, where schemas lacked crucial descriptive information for AI models, leading to incorrect tool selection or hallucinated arguments. The analysis also highlighted that while well-documented, large tool catalogs are possible, they are rare, and maintaining such documentation requires significant discipline. AI
IMPACT Highlights critical gaps in AI agent tool integration, emphasizing the need for better documentation and schema design for improved reliability.
RANK_REASON Analysis of a tool's performance on a set of servers. [lever_c_demoted from research: ic=1 ai=0.7]
- airbnb
- Airtable
- @apify/actors-mcp-server
- brave-search
- exa
- figma-developer-mcp
- firecrawl-mcp
- GitHub
- google-maps
- mcpgrade
- Model Context Protocol
- MongoDB
- Notion
- perplexity-ask
- @shopify/dev-mcp
- slack
- Stripe
- Supabase
- tavily
- todoist-mcp-server
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
