Urgent.News

What's breaking now, across hundreds of outlets.

Tech

MCP standardized tools. Now we need the infrastructure around them.

Developers now have the ability to connect various tools and systems to a common protocol, enabling AI models to interact with them more seamlessly. This shared protocol allows clients such as Claude Code, Cursor, Codex, VS Code, and others to communicate effectively with different tools and services. Vendor-hosted endpoints are available for popular platforms like GitHub, Linear, Notion, and more, making the integration process more straightforward.

While the protocol simplifies the initial connection, the real challenges lie in building the necessary infrastructure to support reliable, secure, and efficient interactions at scale. Developers find themselves reinventing the wheel by creating authentication, deployment, connection management, retries, logging, secret handling, health checks, access control, simple registries, caching, and rate limiting - all of which were previously built independently for each application.

The protocol's success in standardizing the message format has led to a noticeable duplication of operational concerns across different codebases. Developers repeatedly build similar infrastructure components from scratch, wasting time and effort on tasks that were once automated. This duplication is not unique to MCP and has been observed in previous infrastructure transitions, such as the shift from physical servers to cloud abstractions and the evolution of authentication systems.

As the adoption of MCP grows, the need for more robust and standardized infrastructure becomes increasingly apparent. While local servers connected via stdio provide a simple and intuitive starting point, they quickly become inadequate when scaling to accommodate multiple users, real credentials, and production workloads. Remote or hosted servers offer some solutions but introduce new challenges like authentication, latency, multi-tenancy, data handling, uptime, and vendor dependence.

To address these issues, developers are exploring various approaches, such as optimizing repository context for coding agents and implementing progressive or dynamic tool discovery to minimize the amount of data sent to the model. These efforts aim to improve the efficiency and relevance of the information provided to AI models while reducing the token consumption.

However, the question remains: how should the operational concerns be managed at scale? Should developers continue to build their own infrastructure layers, or should they look for higher-level abstractions that handle these concerns more effectively? The answer is still unclear, and the evolution of the infrastructure around MCP protocols will likely determine the future of AI tool integration and development.

Rewritten by urgent.news; may contain errors — check the original.

Source: stackoverflow.blog

Read the original at stackoverflow.blog →

Report a problem with this story

More in Tech

More from Thursday 15 October →