Back to feed
Dev.to
Dev.to
7/22/2026
I lint-scanned 36 popular MCP servers. A third of them are failing your agent.

I lint-scanned 36 popular MCP servers. A third of them are failing your agent.

Short summary

The author built mcpgrade, a Lighthouse-style scanner for MCP servers, and tested 36 popular ones. A third scored D or F — including official servers from MongoDB, Notion, and Airtable — almost entirely because parameters lack descriptions, leaving models to guess argument formats and semantics. The highest-leverage fix for agent reliability is simply adding .describe() to every parameter in your schema.

  • •36 popular MCP servers scanned; 11 scored D/F including official MongoDB, Notion, Airtable servers
  • •Root cause: parameters lack descriptions — schemas generated from zod/OpenAPI without .describe()
  • •Fix is low-effort, high-impact: document every parameter to dramatically improve agent tool-selection accuracy

Generated with AI, which can make mistakes.

#ai-tools#ai-agents#open-source
Read full article at Dev.to

Is this a good recommendation for you?

Comments

Failed to load comments. Please try again.

Explore more

Your MCP Server's Client Is a Language Model. Write the Contract Like You Mean It
Dev.to
The original title is "MCP Scorecard: 4-Layer Pre-flight Check Before You Publish a Server"
Dev.to
I scanned my MCP setup and it scored 0/100. Here's what was wrong.
Dev.to
MCP Tool Descriptions Can Hijack Agent Behavior — A Static Scanner to Detect Injection
Dev.to