MCP Playground
Wiretap the protocol traffic between your agent and its tools.
MCP is 'just JSON-RPC' in theory but invisible in practice; the wire-tap makes every tool call a literal, readable line of protocol traffic — the same view a security reviewer would actually use.
What goes in, what comes out
Learner connects an agent to a set of MCP servers (some well-scoped, one deliberately over-permissioned) and a live wire-tap panel shows the raw JSON-RPC traffic — every list_tools, call_tool, and resource read — as it happens, not just the agent's summary of it. A permission-scope diff then compares what each tool's manifest claims against what it actually did during the run, surfacing overreach in red before it becomes a habit.
A task that requires tool use, and a chosen set of MCP servers (including at least one intentionally over-scoped trick server in the security scenario).
A raw protocol traffic log, a permission-scope red/green diff, and a scored checkpoint for catching the overreach before calling it.
Which MCP servers are attached, read-only vs. read-write scoping per server, and whether the malicious-server scenario is enabled.
In-module lab for AIE-303 (SHOULD tier, feeds the Integrator badge); reused inside Red-Team Dojo as the MCP-specific attack surface.
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.