1. Connect a source
Read supported AI configuration from selected GitHub repositories, or send runtime events from your agents. Start with a small, owned scope.
Know what your AI can access.
Your field guide to AI inventory, access, evidence, and security testing. Start with one repository. Follow a finding back to its source. Know what has actually been checked.
From discovery to a defensible decision
Section titled “From discovery to a defensible decision”2. Understand the evidence
Explore agents, tools, models, identities and destinations. Distinguish what configuration declares from what telemetry observes or a live check verifies.
3. Verify and follow through
Triage findings, run scoped assessments, record decisions, and rerun discovery after remediation. Review coverage before drawing conclusions.
Choose your next step
Section titled “Choose your next step”I am new to ModelFenceSign in, choose an organization, and complete your first investigation.
I want to connect my agentsGitHub configuration discovery and the runtime event ingestion API.
I am investigating a findingRead severity and evidence, trace access, and record a resolution.
I administer a workspaceRoles, access, policy changes, audit logs, and privacy.
What ModelFence does today
Section titled “What ModelFence does today”ModelFence analyzes supported MCP configuration, Claude Code project settings, and AI agent usage in GitHub Actions. It accepts instrumented runtime events and can run an explicitly authorized remote MCP authentication probe. Policies report findings in Monitor mode; they do not block traffic or deployments.
An empty findings list is not a clean bill of health. Unsupported capabilities, missing integrations and incomplete scans remain visible in Coverage. See capabilities and limitations before deciding what a result proves.