MCP In Practice (how MCP is being used) (asimovaddendum.substack.com)

Anthropic’s Model Context Protocol (MCP), launched in November 2024, is rapidly becoming a foundational standard for connecting AI models to diverse tools and data sources via defined servers and clients. MCP enables model-agnostic interoperability, allowing AI agents to read context from and write actions to various platforms, from GitHub repositories to design tools like Figma. This protocol underpins hundreds of AI pipelines, powering applications in software engineering, web automation, and database querying, while promoting decentralized AI ecosystems through open APIs and shared memory standards. An extensive analysis of nearly 2,900 MCP servers reveals a highly concentrated ecosystem where a handful of key servers dominate usage, particularly those enabling agentic browsing and software development functions. Notably, top MCP servers support both read and write operations, indicating a trend toward AI agents not only consuming information but actively interacting with user environments—editing files, automating tasks, or navigating web content autonomously. This rise in agentic browsing reflects gaps in API availability and restrictions, leading developers to rely on browser automation to facilitate AI interactions with web services despite its inefficiencies and challenges like anti-bot protections. The study highlights critical implications for the AI/ML community: concentrated MCP dependencies create potential bottlenecks and points of control, underscoring the importance of local processing capabilities and multi-API support to mitigate gatekeeping risks. MCP’s role as a protocol shapes AI market dynamics by embedding interoperability and permission rules into the network itself, fostering a more open and composable AI ecosystem that can evolve beyond restrictive, single-provider API paradigms.