Playground · the whole loop, live

Run Attestral. Right here.

Every line below is real captured CLI output, streamed like a live session. Click a command to run it, or hit run the full loop and watch a design go from scan, to a reachable attack path, to a runtime policy, to caught drift (the running system diverging from the reviewed design). No install needed to look; pip install attestral when you want it on your own repo. In the output, (model) marks a fleet-level finding that spans several servers, a lethal trifecta means private data + untrusted input + a way for data to leave, and AARS is the Agentic AI Risk Score on a 0–10 scale.

Output is captured from the real CLI against the repo's example fixtures.

It's not just fixtures. Point it at a real repo.

The playground runs on the example fixtures, but the tool is the same one you install. Here is what a first scan actually returns on real, public MCP servers, in about a second, with zero configuration.

18
github-mcp-server
findings in 0.3s. Unauthenticated remote server, secrets in env, one name resolving to two launch targets.
9
mongodb-mcp-server
findings in 1.0s. High and medium issues across five modeled components.
208
aws-mcp
findings in 1.0s, five critical. A cloud-heavy repo where the model surfaces the real signal in the noise.

Run it on your own: pip install attestral && attestral scan .. It reads only what you point it at, runs on your laptop, sends nothing anywhere.