Early developers combined services like Google Maps and Craigslist to create novel applications, sparking the API economy.
Mashups declined due to the high maintenance burden of changing APIs and the lack of long-term commitment for side projects.
The Model Context Protocol (MCP) is an open standard that acts as a universal adapter for AI agents to interact with various APIs and tools.
MCP's open-source nature has led to a massive, community-driven ecosystem of servers and SDKs in languages like TypeScript, Python, and Rust.
An AI agent uses a natural language prompt to combine Google Maps and developer tools to build an interactive restaurant map in seconds.
An AI agent uses MCPs for Figma and developer tools to automatically generate a functional website from a design file.
An AI agent automates the entire setup for a complex project by reading instructions, cloning repos, installing dependencies, and configuring containers.
An MCP server is built by defining tools with clear names, descriptive purposes, parameters, and an execution body using an SDK.
The agentic loop is the process where an agent interprets a prompt, selects the best tools via an LLM, executes them, and repeats until the task is complete.
Developers can use community directories like Glam AI and Post MCP to discover new MCPs and check their quality and security via report cards.
The Goose AI agent is now supported by an independent team and a grant program to foster open-source community contributions.
24:12Brad Axen
39:11Thomas Dohmke
25:32Viktoria Semaan
29:03Jörg Neumann
24:22Philipp Schmid
19:20Jan Curn
25:14Ivan Burazin
23:33Oliver Will
Jobs that call for the skills explored in this talk.
