AXguard exposes security engines to AI coding agents over MCP (Model Context Protocol) on local stdio. No remote MCP URL or AwareXone hosting is required.

​
Install and verify

Point --project / AXGUARD_ROOT at the workspace you want analyzed. The host command must resolve to the same environment where the MCP extra is installed.

​
Tool catalog (31)

Prefer axguard_security_review unless you need a focused follow-up.

​
Primary and understanding

  • axguard_security_review
  • axguard_get_project
  • axguard_get_application_model
  • axguard_get_attack_surface

​
Scan and model

  • axguard_scan
  • axguard_audit
  • axguard_threat_model

​
Data flow

  • axguard_trace_flow
  • axguard_find_taint_paths
  • axguard_find_sensitive_flows

​
Findings

  • axguard_list_findings
  • axguard_get_finding
  • axguard_verify_finding
  • axguard_verify_fix

​
Evidence

  • axguard_get_evidence
  • axguard_get_evidence_chain
  • axguard_get_counter_evidence

​
Attack paths

  • axguard_find_attack_paths
  • axguard_get_attack_path
  • axguard_explain_attack_path

​
Security Twin

  • axguard_get_security_twin
  • axguard_compare_security_twin
  • axguard_what_if
  • axguard_blast_radius

​
Security Memory

  • axguard_get_security_memory
  • axguard_get_security_history
  • axguard_find_regressions

​
Investigation and predictive

  • axguard_investigate
  • axguard_get_investigation
  • axguard_predict_security_risks
  • axguard_analyze_change_risk

​
Approvals

Annotations (readOnlyHint, …) are MCP hints. AXguard’s policy engine enforces approvals separately. Unknown tools default to requiring approval.

​
Agent skill

Teach agents when to call MCP with skill axguard-security — see Agent skill. Client-specific setup: AI agents.