Samurai Security
Run security tests
Run Samurai Security's live security test on a saved server from the web app or the Guardian CLI, and read the results and severities.
This guide runs a security test on one of your saved MCP servers and shows how to read the result. To learn what the test checks and what you need first, see What security tests do.
Run a test in the web app
You need to be an Admin or a Security member, on a plan that includes Samurai Security.
- Save the server on the MCP Servers page if you haven't already, enter its credentials yourself, and mark it as production.
- In the side menu, open Samurai Security, then Security tests.
- If an Admin hasn't accepted the Terms yet, an Admin selects Accept Terms. Read the points first: they cover everyone on your team.
- Review the Security test plan on the page. It lists every test, what it applies to, exactly what is sent, the findings it can report and whether it sends an attack payload.
- Under Server to test, choose your server. The list shows only http, production servers whose credentials you saved yourself.
- Select Run security test and keep the page open. It can take a few minutes.
If the button isn't available, the page tells you why: your plan doesn't include Samurai Security, the Terms aren't accepted, your role can't run tests, or no saved server qualifies.
Read the results
The result shows the Status (Complete or Failed), the Highest severity, the number of Findings and when the test ran. A failed run could not finish and has no results.
- Results summary: a matrix with your server and its tools down the side and the attack classes across the top. Each cell is failed, passed, could not be verified, or not applicable or skipped. The server row comes first, then the tools that failed, then the rest. Select a cell to jump to that tool's details.
- Results detail: each tool with its Attack class, Test, Status (Failed, Passed, Unverified or Skipped), Severity, Detail and a Suggested fix for every failure.
- Passed checks, Could not be verified and Skipped tools list the rest, each unverified check with the reason.
Severities are Critical, High, Medium, Low and Info. Fix the highest first, then run the test again to confirm. Earlier runs stay under Security test history, which anyone on the team can open.
Run a test from the Guardian CLI
The Guardian CLI asks Guardian's cloud to run the same test on a saved http server, which suits a CI pipeline. The CLI never connects to your server and sends no URL, header or credential: Guardian uses the credentials you saved with the server.
- Install the CLI and sign in as described in Running tests locally with the CLI tool, or create a token as described in CLI tokens.
- Use a run token created by an Admin or a Security member. A token from another role is refused.
- Make sure an Admin has accepted the Security Terms in the web app. The CLI can't accept them.
- Run the test with the saved server's exact name:
guardian security-test SERVER
The CLI shows what it will send and asks you to confirm. In a CI job, pass --yes because there is no terminal, and use --fail-on to choose the severity that fails the job:
guardian security-test SERVER --yes --fail-on high
The command exits with a non-zero code when a finding reaches that severity or when nothing could be verified. The run also appears under Security test history in the web app. Use a token with a short lifetime made just for the job, because a token from an Admin or Security member can start attack-payload runs until it expires or is revoked.