
MCP Misuse: Context Bloat, Output Handling, and Skills

The MCP Security Risks You Can't Afford to Ignore - Ariel Shiftan
Topic 6 of 12
Published Mar 2, 2026 · Clip 20:01–30:52 (10:51)
Ariel addresses the two main complaints against MCP: context bloat from long tool descriptions injected into every session, and wasteful output handling where data passes through the LLM context unnecessarily. He explains these stem from the adoption jungle and lack of education, references Anthropic's "code execution with MCP" blog, and then explains Skills as markdown-based progressive disclosure that partially replaces MCP for simple, local use cases, though MCP remains necessary for remote access, authentication, and statefulness.