Authority silo
AI agent builder silo
Builder-oriented agent frameworks, MCP infrastructure and development tools.
This hub connects framework, MCP and coding-agent content without duplicating canonical entity pages.
Why these three sit together
Building an AI agent from scratch touches three distinct layers that are often conflated. A framework (LangGraph, CrewAI, AutoGen) is the orchestration layer — it decides how reasoning steps, tools, and state flow through your application. MCP is the interoperability layer — a standard protocol for exposing tools and context to any compatible model, independent of which framework you use. An agent is the shipped product — the end result someone actually deploys and uses.
A builder evaluating this stack typically needs to answer three separate questions: which framework fits the orchestration pattern, which MCP servers already expose the tools needed, and which existing agents solve adjacent problems worth studying or reusing. Treating these as one undifferentiated "AI tools" category obscures which layer a given evaluation criterion actually applies to — a framework's learning curve and an MCP server's transport security are not comparable facts.
Where to start
If you are choosing a framework, start with the frameworks directory and compare orchestration patterns before capability claims — most frameworks can eventually do most things, but the default patterns differ significantly. If you already have a framework and need tool access, check the MCP server directory for existing, source-linked implementations before building a custom integration. If you are scoping a new agent product, the agent directory shows verified identities for comparable existing products.