The workflow graph
Every agent, tool and MCP hop in n8n, Zapier, LangGraph and CrewAI, mapped and given an owner.
- Every hop mapped
- Owners for every flow
- Unregistered flows flagged
Multi-agent pipelines fail one hop at a time. Zenstra Workflow Security checks each hop against policy and the task the workflow declared, before the next step runs.
Workflows chain agents, tools and MCP calls. An injected instruction at one step travels to the next.
intake-agentAn n8n intake agent receives a customer email carrying an instruction.hand-offThe instruction is passed along from agent to agent, hop after hop.hop 7The flow tries crm.export to an outside email address.hop 7 · crm.export → outside emailEach hop is checked against the task the workflow declared, so one bad step cannot carry the rest of the flow with it.
denyhop.*when outside(workflow.task)denycrm.exportto outside emailholdcore.writefrom no-code flowsallowhandoffintake-agent → triage-agentWorkflow Security sees the whole graph and decides each step before it executes.
Every agent, tool and MCP hop in n8n, Zapier, LangGraph and CrewAI, mapped and given an owner.
Each hop is allowed, held for a person or blocked before the next step executes.
Every hop and decision is on the record, so a failed flow can be replayed and explained.
Self-hosted and cloud workflows with agent steps.
Automations that call AI and write to business systems.
Agents built by business users inside Microsoft 365.
Graph-based agents your engineers build and run.
Multi-agent crews with hand-offs between roles.
Tool calls made from any step in the flow.
No. Workflow Security works with the flows you have; there is no agent rewrite.
The workflow pauses at that step until a person approves or rejects it, and the decision is recorded.
Yes. Each hand-off is checked against the original task, so an injected instruction cannot ride along.