Intel / article

MCP was always a bad idea?

Source: simonwillison.net
Date: Published Sep 20, 2026

Context

's usefulness depends on how an application controls an 's access to other systems. The Model Context Protocol standardizes connections between AI applications, tools and data. Simon Willison argues that its value remains substantial when developers need to limit which services an assistant can reach and keep access credentials out of the agent's direct control.

In his comment, Willison also points to user authentication and audit logging. His argument concerns the surrounding application: a terminal agent with unrestricted access can call services directly, while a more constrained assistant benefits from managed connections. That explains why terminal agents' capabilities alone cannot settle whether MCP is useful.

Key takeaways · AI-distilled

Willison concedes MCP is largely unnecessary for full autonomous terminal agents (Claude Code, Codex, Meta Muse, OpenClaw) that already have unfettered API access — they can just call APIs directly.
But he argues MCP earns its keep for less permissive setups needing scoped access control, credential handling that keeps API keys away from the agent itself, a connect/authenticate UI for end users, and strong audit logging.
His conclusion: calling MCP obsolete conflates 'unneeded by fully autonomous coding agents' with 'unneeded everywhere' — ignoring the broader class of applications that need those controls.

Terms in this piece · Glossary

MCP — The Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.
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.

Read the source simonwillison.net