Mode A is the project-owner path for AgentVeil. It is for teams using Cursor, Claude, MCP servers, CrewAI, GitHub Actions, or similar automation inside a project and asking: what can these tools do, which actions are risky, and which paths should be routed through AgentVeil?
AgentVeil is one action-control system. Identity, delegation, approvals, receipts, and reputation signals are decision inputs and evidence mechanisms, not separate public product promises.
Lurkr is a local pre-runtime scanner for risky AI-agent capabilities.
pip install lurkr
lurkr scan --path ./your-agent-projectTreat the report as a triage input before wiring runtime controls.
The public routed path available today is MCP Proxy:
pip install agentveil-mcp-proxy
agentveil-mcp-proxy init --quickstart-filesystem ./sandbox
agentveil-mcp-proxy doctor --full
agentveil-mcp-proxy runMCP Proxy controls only MCP calls routed through the proxy. Actions outside the proxy are not classified or logged.
For backend-connected SDK paths:
- Register or load an agent identity.
- Receive a delegation receipt from the workflow owner.
- Call
controlled_action(...)before the sensitive action. - If approval is required, route approval to the principal and resume after approval.
- Store receipt text or a proof packet for later review.
Relevant guides:
| Capability | Status |
|---|---|
| Lurkr pre-runtime check | Available as lurkr |
| MCP Proxy routed MCP calls | Available as agentveil-mcp-proxy |
| SDK controlled-action calls | Available for integrated application paths |
| Local policy CLI | Planned |
| Credential custody boundary | Preview/design-partner pattern |
| Egress boundary | Preview/design-partner pattern |
| API gate boundary | Preview/design-partner pattern |
Public claims should track this table. If a capability is listed as planned or preview, do not treat it as shipped behavior.
Agent identity, reputation credentials, delegation receipts, credential helpers, and optional framework adapter modules remain available as advanced protocol primitives when their framework dependencies are installed. See Agent Network (Advanced) when you need those primitives directly.