Skip to content

Capabilities and limitations

Use this page to decide whether a ModelFence result answers your question. The application’s Coverage view is the source-specific record of what actually ran.

  • GitHub App discovery of supported repository configuration.
  • MCP configuration, Claude Code project settings and supported GitHub Actions AI agent analysis.
  • Passive findings with source evidence and stable correlation.
  • Explicitly authorized remote MCP authentication probing.
  • Runtime JSON ingestion, token rotation, allowed-attribute processing and observed relationships.
  • Organization-scoped inventory, findings, graph, data flows, assessments and reports.
  • Owner, Admin, Analyst and Viewer roles, with server-side permission checks.
  • Monitor-only detection policies with version history and a dry-run workflow.
Area Boundary
Cloud AI service discovery No full Azure AI Foundry, Amazon Bedrock or Vertex AI inventory.
Framework code No general permission analysis of arbitrary LangChain, Semantic Kernel or Agents SDK implementations.
Model supply chain No evaluation of model weights, provenance or serialization risks.
Prompt injection and model red teaming No general adversarial model testing harness.
Enforcement No in-path blocking of tools, traffic or deployments.
Payload capture Prompt, completion and request bodies are not accepted as supported runtime content.
Collector package No packaged OpenTelemetry Collector exporter or receiver; send supported JSON events.
Invitations and SSO groups No self-service email invitations or automatic group synchronization.
Scheduled reporting No report scheduler.
SecureObs Read-only payload preview; no live integration.

Inventory saved views are browser-local; bulk operations and a column chooser are not provided. Access Explorer has focus, hop limits, filters, an inspector and a table, but no automatic path-between-two-nodes selection. Settings does not expose retention or subscription management.

Real sign-in, GitHub access, active-assessment entitlements and runtime ingestion depend on the operator’s configuration and your organization’s permissions. A feature appearing in the source code does not mean it is enabled for every deployment or organization.

No finding, dashboard or badge is a compliance certification or a guarantee that an environment is free of risk.