AI agents are finally getting a knowledge infrastructure
Official skills and MCP servers from Google, JetBrains, Microsoft and OpenAI are turning agents from chatbots into engineering tools.
It looks like AI agents are finally getting a proper knowledge infrastructure. It used to be all about one model and a prompt. Now the platforms themselves are shipping official skills and MCP servers, so an agent can rely on up-to-date documentation, platform patterns and ready-made workflows instead of “whatever it remembers”.
A few good examples:
Google MCP — an official set of MCP servers from Google: Developer Knowledge API, BigQuery, GKE, Maps, Firebase, Google Cloud and more. This is already a way to plug live Google services and docs into your agent.
Developer Knowledge API / MCP Server — a separate case from Google: access to current developer documentation, including developer.android.com, over MCP.
Android Skills — an official repository from Android: modular skills for LLM/agent workflows built on Android development best practices.
Microsoft / C# Agent Skills — Microsoft is moving the same way: dedicated skills and guidance for C# / .NET, so the agent writes code for the Microsoft ecosystem instead of “something vaguely similar”.
Kotlin Agent Skills — JetBrains is developing the same approach for Kotlin, KMP and Android.
GitHub Spec Kit — not MCP, but an important layer too: a toolkit for spec-driven development where the agent works through a spec, a plan and an implementation.
iOS / Apple is a special case
The contrast is telling: there’s still no official MCP server or skills repository from Apple itself. But the community and vendors are actively filling the gap.
- OpenAI Codex — iOS / SwiftUI skills. OpenAI shipped a whole collection of skills specifically for iOS development: SwiftUI Pro (reviews for modern APIs, accessibility, performance), Liquid Glass expert (adapting to the new iOS 26 APIs), SwiftUI performance, Swift concurrency expert, SwiftUI view refactor, SwiftUI patterns. Plus a general skills catalog in the repository.
- XcodeBuildMCP (Sentry). An MCP server + CLI that gives the agent full control over Xcode: builds, tests, running on simulators and devices, LLDB debugging, UI automation. Since Xcode 26.3 it integrates natively with Claude and Codex right inside Xcode.
- Sosumi — Apple Docs for LLMs. An MCP server and skill for accessing Apple Developer documentation, the Human Interface Guidelines and WWDC session transcripts.
The main point is simple: the future of AI development is no longer just a model, but a bundle of:
- an agent
- skills
- MCP/tooling
- official sources of knowledge
And this is exactly what’s gradually turning AI from a “smart chatbot” into a reasonably sane engineering tool.