Use Cases¶

Oktsec solves real security problems in AI agent deployments. These aren't theoretical — they're the attack patterns seen in production multi-agent systems.

Who is this for?¶

Multi-agent teams¶

You have agents coordinating work — a planner, a coder, a reviewer. You need to ensure they can't hijack each other or leak data across trust boundaries.

Multi-Agent Pipeline

MCP server operators¶

You expose tools (filesystem, database, APIs) to AI agents via MCP. You need to control which agents access which tools, and scan every tool call.

MCP Server Hardening

Security-conscious teams¶

Your agents make outbound HTTP requests. You need to ensure they don't exfiltrate credentials, PII, or sensitive data to unauthorized endpoints.

Egress DLP

OpenClaw users¶

You run OpenClaw agents with MCP tools. You need to control which agents call which tools, enforce spending limits, and scan every tool call for threats.

OpenClaw Integration

Common attack patterns Oktsec catches¶

Attack	Rule	Verdict	What happens
Agent A tells Agent B to ignore its system prompt	IAP-001 (critical)	Block	Message rejected, webhook alert
Credentials embedded in agent messages	IAP-003 (critical)	Block	Message rejected, audit logged
Agent tries to extract another agent's system prompt	IAP-004 (high)	Quarantine	Held for human review
MCP tool description contains hidden instructions	IAP-007 (critical)	Block	Tool call rejected
Tool name typosquatting (`read_flie` vs `read_file`)	IAP-012 (high)	Quarantine	Tool call held for review
Outbound HTTP request contains API keys	Aguara credential rules	Block	Egress request blocked
Agent sends data to unauthorized domain	Per-agent egress policy	Block	Request blocked at forward proxy
Same agent triggers 5+ blocks in an hour	History escalation	Block	All future content auto-escalated

How teams deploy Oktsec¶

DevelopmentStagingProduction

oktsec setup
oktsec serve

Zero config. Discovers your MCP servers, wraps them, starts in observe mode. Review activity in the dashboard before enabling enforcement.

# oktsec.yaml
identity:
  require_signature: true
default_policy: deny
agents:
  planner:
    can_message: [coder, reviewer]
  coder:
    can_message: [planner]
    allowed_tools: [read_file, write_file, run_tests]
    blocked_content: [credentials]
  reviewer:
    can_message: [planner]
    allowed_tools: [read_file, search_files]

Enforce signatures, default-deny unknown agents, restrict tool access and content categories per agent.

identity:
  require_signature: true
default_policy: deny
rate_limit:
  per_agent: 100
  window: 60
anomaly:
  risk_threshold: 50
  auto_suspend: true
forward_proxy:
  enabled: true
  scan_requests: true
  blocked_domains: [pastebin.com, transfer.sh]
webhooks:
  - name: slack-security
    url: https://hooks.slack.com/services/xxx
    events: [blocked, quarantined, agent_risk_elevated]

Full enforcement with rate limiting, anomaly detection with auto-suspension, egress DLP, and Slack alerts on security events.