Accident Investigation Prover MCP for AI. Forces NTSB-level rigor into every incident report.

Yes, it must correlate evidence like FDR parameters and CVR transcripts to validate any conclusion. Without these raw inputs, the tool won't proceed past initial checks.

A standard model writes narratives; this MCP forces adherence to NTSB/ICAO methodology. It demands classification via HFACS and requires mapping contributions back through organizational factors.

While designed for aviation, its framework—analyzing evidence chains, causal links, and systemic failure across four levels—can be adapted to other high-stakes technical fields.

The output provides a structured analysis that includes probable causes, contributing factors (NTSB format), and actionable recommendations linked directly back to specific evidence sources.

You simply connect via your preferred MCP-compatible client through Vinkius. The platform manages all authentication and authorization, so you don't need to worry about API keys or complex setup steps.

The tool is built to handle gaps. If the input lacks key evidence, it won't give a conclusion; instead, it generates a report detailing the structural deficiencies and missing links in your investigation.

No. We don't store your proprietary incident data. The MCP processes the information solely for generating the analysis, and all inputs remain confidential to you.

Vinkius handles scaling for high volume usage. While standard use is robust, extremely high-frequency requests can be managed through our enterprise tier options.

The engine maintains a semantic trap list of blame-language signals: 'pilot error,' 'crew error,' 'human error,' 'judgment error,' 'failed to,' 'negligence.' If the LLM uses any of these in the HFACS taxonomy field, the classification is rejected. Instead, the LLM must classify each factor at the correct HFACS level: Level 1 (what the pilot did — skill error, decision error, perceptual error, or violation), Level 2 (what conditions enabled it — fatigue, CRM failure, environment), Level 3 (what supervision allowed it — scheduling, training gaps), Level 4 (what organizational decisions created it — budget cuts, staffing, regulatory gaps). 'Pilot error' is Level 1 only — the investigation must reach Level 4.

Five mandatory sources: (1) Flight Data Recorder — minimum 88 parameters per ICAO Annex 6, with timestamps. (2) Cockpit Voice Recorder — last 2 hours of audio, transcribed with timestamps, correlated with FDR parameter changes. (3) ATC recordings — radar track, clearances, handoffs, weather advisories. (4) Maintenance logs — last A/B/C/D checks, MEL items, deferred defects, AD compliance, component life (TSN/TSO/CSN/CSO). (5) Wreckage analysis — impact signatures, fire patterns, fracture analysis (fatigue vs overload), metallurgical examination. The engine rejects speculative language like 'it appears that' or 'evidence suggests' when hard data from these sources exists.

Because 'improve training' with no addressee has zero accountability. The NTSB model requires each recommendation to name the authority responsible for implementation — the FAA, the operator, the manufacturer, or an international body. Each recommendation must include: what specific action to take, measurable success criteria, a response deadline (typically 90 days), a verification mechanism (audit, inspection, data review), and a direct link to a specific investigation finding. This is how aviation achieved its extraordinary safety record — not through vague wishes, but through tracked, accountable, evidence-linked changes.