MCP: tools from a server
How an agent finds and calls tools on an MCP server, and how the 2026-07-28 protocol asks a person before a risky call.
MCP, A2A, AG-UI and friends, orchestration, context, tracing and testing, with the real protocol code.
For developers who can already build a single agent. You connect agents to tools, to other agents, to interfaces and to payments with the open standards, orchestrate several agents in one graph, decide what goes into each context, trace runs and test them against attacks. The protocol libraries run for real in your browser; you can hold, drop and tamper with every message.
How an agent finds and calls tools on an MCP server, and how the 2026-07-28 protocol asks a person before a risky call.
How one agent finds another through its Agent Card and hands it a task it can follow to the end.
Streaming an agent's work to an app with AG-UI, and letting an agent describe safe interface components with A2UI.
How an agent checks out with a merchant over UCP, and how AP2 mandates bind the payment to what the person signed.
Patterns for several agents in one LangGraph graph: a supervisor that routes, Send for parallel work, subgraphs, and the checkpoint each superstep saves.
Choosing what each model call reads: grep or vector search, offloading large results, a first look at subagents, and memory that lasts across threads.
Recording every model call, tool call and step as a tree of runs, and reading it to find what is slow or broken.
Testing an agent system with cases and checks, attacking it with injection and tampering, and mapping the failures to the OWASP agentic risks.