The Model Context Protocol: Overkill or the Future of Agentic Workflows?

The Model Context Protocol (MCP) has sparked a debate among developers regarding the necessity of its architecture. At its core, the protocol aims to standardize how AI agents interact with data and tools. However, critics argue that the functionality it provides—such as command discovery and permission scoping—has already been solved by the Command Line Interface (CLI) for decades.

This tension highlights a fundamental question in the AI era: are we adding unnecessary complexity to our local environments by introducing a new layer of processes, or is MCP providing a critical abstraction that CLIs cannot offer?

The CLI Argument: Why Another Process?

For many seasoned developers, the Model Context Protocol seems redundant. The primary arguments against the proliferation of MCP processes on a local machine include:

• Permission Scoping: CLI commands can already be scoped with specific permissions, limiting what a tool can do on a system.
• Discoverability: The standard --help flag and man pages provide a robust way for users (and potentially agents) to discover commands and their required arguments.
• Resource Overhead: Running dozens or hundreds of separate MCP processes on a single PC creates unnecessary overhead and increases the surface area for potential system instability.

From this perspective, MCP is seen as adding "extra complexity and moving parts for not much benefit."

Beyond the Local Process: Server-Side MCP

One counter-argument to the "100 processes" concern is that MCP does not necessitate local execution. The protocol is designed to be flexible, allowing for server-side hosting.

As noted by community members, companies like Atlassian are already implementing this approach. By hosting MCP on the server side and utilizing Dynamic Client Registration and OAuth, organizations can provide secure, authenticated access to tools without burdening the user's local machine. This allows for the use of tools like oauth2-proxy and Nginx to wrap open-source MCPs with Single Sign-On (SSO) layers, making them enterprise-ready.

State Management and Auditability

While the CLI is excellent for executing a command and receiving a result, it fails to address the state management required for complex, multi-step agentic workflows.

When an AI agent is performing a task that spans fifty steps, a simple CLI call cannot easily "rollback" the context or recover from a failure at a specific point. This is where the shift toward more structured protocols becomes vital. The need for Git-like branching and snapshots for agent memory is emerging as a critical requirement for making these processes auditable and recoverable, transforming the "100 processes" from a liability into a structured, manageable system of record.

Conclusion

Whether the future involves hundreds of local MCP processes or a centralized server-side architecture, the Model Context Protocol represents a shift in how we think about tool integration for AI. While the CLI remains a powerful tool for human developers, the requirements of AI agents—specifically regarding state, remote authentication, and standardized discovery—suggest that a more formal protocol may be necessary to move beyond simple script execution toward truly autonomous agentic workflows.