Guardful AI

MCP and agent tools

Give an agent explicit policy discovery, estimate and check tools

Example

  1. Proposed agent message

    “Please contact sam@example.test about order 2041.”

  2. aiguard_check_content

    Estimate approved · Email address found · Review

  3. Host enforces the result
    Completed Allow

    safe_to_continue=true

    Permit the next action
    Review
    Stop and request human review

Block, incomplete results and errors also stop the next action; the host also requires continue_allowed=true and required_route=continue

Setup

  1. Install the supplied stdio MCP server
  2. Configure its executable and protected credentials in your host
  3. Discover a policy version and approve the estimate for your content
  4. Check the content and have your host enforce the returned result
Read getting started