MCP and the innovation paradox: Why open standards will save AI from itself
1 min read
Summary
The disruption in AI isn’t bigger models or LLMs but rather the creation of a data standard, something that HTTP and REST created for web applications, which is the Model Context Protocol (MCP), aiding the interface of AI with tools.
MCP was created to solve the issue of bespoke connection to LLMs which would pull users into specific vendor platforms, meaning that it was hard to switch to different/better LLMs in the future.
The advantage of MCP is that it does not require custom code to test AI apps, resulting in much quicker development cycles for AI apps and near-zero switching costs for other models and providers.
The onus is on SaaS providers to provide APIs in order to avoid becoming obsolete, whilst development cycles are speeding up and providers are having to maintain servers to ensure high-quality metadata for the models.
The challenges that have come from the adoption of MCP are around trust, with third-party registries causing the potential for data leaks, and issues with quality and authorisation.