From Legacy Documentation to OSCAL: An MCP-Based Agent Pipeline for Threat-Informed Continuous Compliance in Critical Infrastructure

Critical infrastructure sectors, such as water and energy, rely on legacy Industrial Control Systems (ICS) that are often too fragile for active vulnerability scanning. However, these systems still require rigorous, documented risk management to meet modern compliance standards like NIS-2 and BSI Grundschutz++. This paper addresses the gap between unstructured, legacy system documentation and the structured, machine-readable models required for automated compliance.

The authors propose a non-invasive, eight-phase multi-agent pipeline that converts natural-language system descriptions into NIST OSCAL (Open Security Controls Assessment Language) artifacts. The architecture uses the Model Context Protocol (MCP) to ground LLM reasoning in deterministic, authoritative threat intelligence sources. By forcing agents to query validated APIs (e.g., NVD, CISA KEV, MITRE ATT&CK) rather than relying on internal model knowledge, the pipeline minimizes hallucinations and ensures that all vulnerability data is verifiable.

The pipeline was evaluated using a synthetic, evidence-based water utility scenario. The system successfully generated schema-valid OSCAL System Security Plans (SSP) and Security Assessment Reports (SAR). Key performance metrics included a 0.90 CVE recall and a perfect 1.00 D3FEND recall.

A critical insight of this research is the shift in error profiles. While the use of MCP eliminates hallucinations during the retrieval phase, the pipeline remains susceptible to "semantic hallucinations" during the initial entity extraction phase (Phase 0). If an asset is incorrectly identified from the source documentation, the subsequent deterministic pipeline will accurately retrieve vulnerabilities for the wrong asset. The authors argue this is a favorable trade-off: it makes the remaining risk visible and verifiable, allowing human operators to perform time-efficient manual reviews rather than hunting for phantom vulnerabilities.

This approach provides a pathway for critical infrastructure operators to achieve continuous compliance without risking the availability of sensitive operational technology. By automating the generation of OSCAL artifacts, the framework reduces the administrative burden of audits while providing a transparent, traceable audit trail. The use of a "safety-first" heuristic for vulnerability prioritization ensures that limited defensive resources are directed toward the most urgent threats, bridging the gap between IT-centric security scoring and the unique safety requirements of industrial environments.