Agent Control (ZASTAPNIK) — the control plane for AI agents and MCP
The newest attack surface is the AI agent. Zastapnik is the action layer that governs agents and MCP tools — tool-poisoning detection, agent-as-identity, and reachability analysis, all on one security graph.
64
Active Agents
211
MCP Tools
3
Poisoned
The headline proof
One sequence no point tool can run
This is the moat in a single chain — and why agent security belongs on the graph, not in a silo.
- 1
Zastapnik finds a poisoned MCP tool on a live agent.
- 2
The graph shows that agent can reach PII via an over-scoped token.
- 3
The contradictions engine surfaces it as a single finding.
- 4
Jadro explains the chain in an auditable reasoning trace that never left the environment.
Capabilities
What Zastapnik does
Three capabilities that bring AI agents and MCP under the same security graph as everything else.
Tool-Poisoning Detection
Catch malicious or drifted MCP tools before an agent runs them
Continuously inspects MCP servers and tool definitions for prompt injection, instruction smuggling, and silent schema drift on live agents.
Agent-as-Identity
Treat every agent as a first-class non-human identity
Each agent gets a tracked identity, scoped credentials, and a full action audit trail — shared with Nadzor's NHI graph.
Reachability Analysis
Know exactly what data and systems an agent can touch
Resolves the real blast radius of an agent through tokens, tools, and the security graph — including transitive paths to PII and secrets.
Architecture
The action layer on the security graph
Zastapnik acts on agents and MCP, then writes everything it learns back to the one graph all ten modules share.
Reads from
Nadzor
Identity & NHI signals
ProtivAI
Model-layer risk signals
Podatoci
Data sensitivity & exposure
Odbrana
Vulnerability findings
Zastapnik
Action Layer
Emits to graph
Agent Identities
Scoped, tracked, audited
Tool-Poisoning Findings
MCP drift & injection
Reachability Edges
Agent blast radius
Use Cases
For organizations deploying real agents
Zastapnik serves teams putting AI agents into production — where one over-scoped tool can reach everything.
AI-Forward Enterprises
- Govern internal copilots
- MCP server allow-listing
- Agent action audit for compliance
Platform & DevEx Teams
- Safe MCP rollout at scale
- Per-agent least privilege
- Drift detection on tool catalogs
Security Teams
- Agent blast-radius reviews
- Poisoned-tool incident response
- Shadow-agent discovery
MSSPs
- Multi-tenant agent oversight
- Client-ready agent risk reports
- Repeatable MCP hardening
Trust
Governed, audited, sovereign
Agent security is high-stakes. Zastapnik is built with authorization and auditability at the core.
Authorized Action
Agent interventions run under explicit authorization with full, immutable audit trails.
Complete Visibility
Every agent, tool, token, and action is discoverable — no shadow agents, no blind spots.
Least Privilege
Per-agent scoped credentials, continuously checked against real reachability.
Sovereign Option
Pair with Jadro to keep agent reasoning fully on-prem or air-gapped. Zero egress.