MCP Debate: Token Tax, Context Bloat, and What Devs Can Do

A recent debate on maharship.com highlights how the Model Context Protocol (MCP) introduced in 2024 has led to context bloat and a "token tax" on modern models. The post argues that MCP was never suitable for contemporary models and is causing developers to reconsider its use.

The controversy sparked a heated discussion on Hacker News and X, with over 165 points and 100+ comments. Some developers argue for scrapping MCP, while others defend it for its auditability and sandboxing benefits. The debate centers on whether the added cost of MCP justifies its use in today's computing environment.

The core issue is that tools and MCP consume processing context, which incurs costs. Developers are forced to consider the trade-offs between using MCP and relying on plain command-line interfaces. The challenge lies in balancing the need for auditability and sandboxing against the desire for simplicity and efficiency.

To address the problem, developers recommend three steps:

1. Measure the context your tools consume. Tokenize tool definitions and system prompts to estimate token usage. This helps identify which tools are most costly and when they are actually needed.

2. Load tools conditionally. Only include the tools necessary for a specific task, rather than loading the entire set on every call. This modular approach reduces the overhead and makes it easier to manage different model providers and keys.

3. Implement a cost log alongside the tool log. Track input tokens to understand the true cost of each interaction and make informed decisions about which tools to use and when.

Ultimately, the MCP debate is not about whether the protocol is valuable but about how its overhead impacts developers. By measuring and optimizing tool usage, teams can mitigate the negative effects of MCP while still benefiting from its auditability and sandboxing features.