Question 1

How do I add it to my application?

Accepted Answer

Use the API or MCP to submit content before it reaches the agent, then apply the verdict in your delivery logic. You can try supported targets in Agents → Context in the dashboard. The docs include inbound email and message examples.

Question 2

What can it scan?

Accepted Answer

Supported targets include public web pages, text files and PDFs, raw email, message text and links, agent skills, public GitHub MCP repositories, and supported registry packages. Each target has its own checks and limits.

Question 3

What do the scores and verdicts mean?

Accepted Answer

Results include scores, confidence, and a safe, caution, suspicious, or dangerous verdict. New lookups can return a preliminary identity score while a deeper scan finishes. Request a completed result when your decision needs the full assessment.

Question 4

Does it block content for me?

Accepted Answer

Your application decides what to do with the verdict. You can pass approved content through, request a review, or discard and alert. Run the check before delivering input to the agent.

Question 5

Can I send personal or health information?

Accepted Answer

PII and PHI use cases are excluded. Classify and filter content before sending it to Context Guardrails. Runtime Guardrails can evaluate separate local rules where a coding agent runs.

Question 6

Does it scan attachments and live MCP servers?

Accepted Answer

Message media and attachments, and email attachment contents, are outside coverage. MCP scans inspect public GitHub repository contents without connecting to or executing a live server. A verdict covers the content inspected, not omitted material.

Screen what your agent reads

Message screening

Page and file checks

Checks before installation

Email before agent delivery

Documents before agent use

more reading

Check an input before delivery