MCP Doesn't Suck. Your Agent Does. — Jan Čurn, Apify
Source
youtube.com
Author
AI Engineer
Date
Context
An AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → can use MCP tools without loading every tool's full instructions at the start. This talk argues that excessive tool descriptions are often a problem in the software surrounding the model, rather than an unavoidable property of MCP, the protocol connecting agents with external tools.
Apify's open-source mcpc client provides a concrete alternative. Its README describes listing each tool by name and a one-line description, searching tools with a grep command, and fetching one tool's full schema only when needed. Its command-line interface still communicates with MCP servers. The README presents this progressive discovery as a way to reduce token use and context windowThe maximum amount of text a model can consider at once — its working memory for the current conversation or task.Full definition → clutter; no evaluation results are claimed here.
Key takeaways · AI-distilled
Čurn argues most 'MCP is dead' complaints trace to naive clients that load every MCP tool into context up front, a agent harnessThe scaffolding around a model that turns it into a working agent — the loop, the tools it can call, and the rules for when to stop.Full definition → bug rather than a protocol flaw.
He names three fixes the industry has landed on: sub-agents, progressive tool discovery, and Code Mode, which works because models are better at writing code than calling tools.
His split: CLIs feel better to agents and suit local access, while MCP wins for remote, authenticated access.
Apify's open-source mcpc is a universal CLI client for MCP covering stdio and remote servers, OAuth, persistent sessions, grep-based tool discovery, async tasks and x402A payment protocol built on HTTP's long-unused 402 status code, letting a server charge per request so software agents can pay without an account.Full definition →. Early Connector Evals compare raw MCP, CLIs and mcpc.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
agent harness — The scaffolding around a model that turns it into a working agent — the loop, the tools it can call, and the rules for when to stop.
x402 — A payment protocol built on HTTP's long-unused 402 status code, letting a server charge per request so software agents can pay without an account.