Companion videos

Designing MCP tools that LLMs actually use correctly — companion videos

The article is about the difference between an MCP tool that lights up the inspector and one a model actually picks up at the right moment — names, descriptions, parameter shapes, error messages, what the tool returns, how it composes with other tools. These two Anthropic sessions are the most direct on-camera treatment of that surface from the people who design Claude's own tools.

Primary pick

18:58

Building more effective AI agents

Anthropic

Anthropic engineers walking through what they actually changed when their multi-agent systems were misusing tools — collapsing endpoints, returning names instead of IDs, leaning on MCPs and Agent Skills instead of stuffing more tools into the system prompt. Maps point-for-point onto the article's checklist for tool descriptions and return-shape design.

What you should get from this: Improve tool design so agents select the right action with the right parameters.

Watch or know first: Know the difference between a tool schema, tool description and agent instruction.

AI Expert note: The examples are Anthropic-specific, but the design lesson is stable: reduce tool ambiguity, return useful identifiers and test tool choice on real tasks. Re-check current Agent Skills and MCP guidance before copying any product-specific pattern.

Open video page

Also worth watching

29:27

Prompting for Agents | Code w/ Claude

Anthropic

Hannah Moran and Jeremy Hadfield from Applied AI walking through how to phrase tool calls and agent prompts on a real Pokemon-playing agent — when to push behavior into the system prompt versus the tool description, what the model needs to know about each tool's preconditions. Useful immediately after you write your first MCP server and find Claude calling it in unexpected ways.

What you should get from this: Decide which behavior belongs in the system prompt, tool description or tool precondition.

Watch or know first: Have built or inspected at least one tool-using agent.

AI Expert note: Treat the demo game as a teaching surface, not the point. The transferable lesson is prompt/tool contract design; validate any production prompt with eval cases that include wrong-tool and missing-parameter failures.

Open video page