Tool X-Ray
Rewrite one sentence, watch the AI agent stop guessing.
Literalizes 'an MCP server is an API product' -- the learner edits words, not code, and watches an agent's behavior change in real time, proving schema quality is the product.
What goes in, what comes out
Learner is shown two versions of the same MCP tool description -- vague and well-specified -- and watches an agent attempt to call each: the vague version produces wrong parameters and retries, the well-specified one succeeds first try. An X-ray toggle exposes the raw schema fields (name, description, param types, scopes) for live editing and re-testing.
Edited tool-description text.
The agent's actual tool-call attempt (correct/incorrect parameters), retry count, success/fail, and a scope-creep warning if the tool is over-privileged.
Tool-description editor (freeform); scenario picker (file read, calendar write, payment action, to vary blast-radius stakes); retry-count display.
Demo/lab companion inside AIE-303; lead magnet for engineers building their own MCP servers.
Go deeper, elsewhere
Hand-picked public explainers and open tools that complement this one — always optional, never required, never graded.
Concepts click when you open the machinery.
Three labs are already live and free — the same hands-on style this playground brings to its module.