Skip to main content
bash TV

MCP Doesn't Suck. Your Agent Does. — Jan Čurn, Apify

AI Engineer

5.1K views2 Oct 2026

YouTube

Everyone's dunking on MCP. The protocol isn't the problem. Your agent harness is. Jan Čurn, founder and CEO of Apify, goes through the 'MCP is dead' backlash and shows that most of the complaints, like context bloat and token burn, come from naive client implementations, not the protocol. He covers the three fixes the industry has landed on (sub-agents, progressive tool discovery, Code Mode), explains why CLIs feel better to agents, and makes the case for using MCP for remote access and a CLI for local access. Then he demos mcpc, Apify's open-source universal CLI client for MCP, and shares early results from Connector Evals comparing raw MCP, CLIs and mcpc. In this talk: • Why loading every MCP tool into context up front is a harness bug, not a protocol flaw • Progressive tool discovery and Code Mode, and why models are better at writing code than calling tools • Why CLIs win locally and MCP wins for remote, authenticated access • A live demo of mcpc: stdio and remote servers, OAuth, persistent sessions, grep-based tool discovery, async tasks and x402 SPEAKER Jan Čurn, Founder & CEO, Apify LinkedIn: https://www.linkedin.com/in/jancurn X: https://x.com/jancurn GitHub: https://github.com/jancurn LINKS Apify: https://apify.com mcpc (GitHub): https://github.com/apify/mcpc Introducing mcpc: https://blog.apify.com/introducing-mcpc-universal-mcp-cli-client/ Apify MCP server: https://github.com/apify/apify-mcp-server CHAPTERS 0:00 Intro 0:12 MCP doesn't suck, your agent does 0:42 How MCP took over the AI world 1:32 The MCP hate wave 2:32 The real complaint: context bloat 3:27 It's the harness, not the protocol 3:46 Fix #1: Sub-agents 4:41 Fix #2: Progressive tool discovery 5:41 Fix #3: Code Mode 7:16 Why agents love CLIs 9:35 Where CLIs fall short 10:15 MCP for remote, CLI for local 10:45 Introducing mcpc 12:14 Demo: local and remote MCP servers 14:49 Demo: progressive discovery with grep 15:24 Demo: async tasks 16:28 x402 support 16:58 Connector Evals: MCP vs CLI vs mcpc 18:18 MCP + CLI is the answer Recorded at the AI Engineer World's Fair 2026 in San Francisco. Subscribe for more talks from the engineers building with AI. AI Engineer: https://ai.engineer YouTube: https://www.youtube.com/@aiDotEngineer X: https://x.com/aiDotEngineer LinkedIn: https://www.linkedin.com/company/aidotengineer/ #MCP #AIAgents #AIEngineer

Join the discussion

Sign in to join the discussion

Sign in