A comment posted on Hacker News argues that the Model Context Protocol (MCP) remains useful for operating "controlled" agents. While some fully networked coding agents — such as Claude Code, Codex, Meta Muse, or OpenClaw — can call APIs directly, that capability does not eliminate the practical benefits MCP provides in many deployment scenarios.
What MCP provides
The commenter highlights four operational areas where MCP makes development and operations easier:
- Fine-grained control over external service access: MCP lets operators specify exactly which external services an agent may reach.
- Authentication handling that protects secrets: the protocol supports authentication flows that avoid giving the agent direct access to API keys or other sensitive credentials.
- A sensible UI for connecting services: an MCP-based setup simplifies building user-facing interfaces that let end users connect and authenticate additional services without sharing credentials directly with the agent.
- Strong audit logging: MCP enables detailed operational logs so it is possible to track what resources and services an agent used, which aids compliance and incident investigation.
Why calling MCP obsolete is misleading
The comment's point is that concluding MCP is unnecessary because some agents can make direct API calls misses the broader operational requirements many applications have. MCP is less about replacing the ability to call APIs and more about providing a mediated surface and governance model that improves security, transparency, and user experience.
Conclusion
In short: direct API-capable agents exist, but MCP remains valuable for deployments that require tighter control, safer handling of credentials, user-friendly service connection flows, and comprehensive auditability. Dismissing MCP as obsolete overlooks these practical operational benefits.
