Most forensic schedule consultancies are still emailing PDFs. CPP is built natively on Anthropic's Claude Skills, Model Context Protocol, and the Claude Agent SDK — the stack that didn't exist twelve months ago. Sixteen production skills, thirteen tools exposed as a hosted MCP server, agent-orchestrated multi-step analyses on real XER files.
The engine is the easy part. The hard part is reading what it tells you — that part still runs through the analyst.
Most consultants either treat AI as a smart Google with no tooling integration, or sell static SaaS dashboards with no AI agency. CPP is in a third category: a forensic analyst who built his own toolkit and made the whole thing agent-native.
Anthropic's framework, released in late 2025, for packaging specialized AI capabilities. Each Skill bundles executable code, methodology references, and trigger documentation that Claude loads only when the task demands it.
CPP runs sixteen of them in production — every analysis methodology a forensic delay claim has to survive, written by a working analyst.
An open standard, also from Anthropic, that lets any AI assistant call external tools. Once a server is published, Claude Desktop, claude.ai, Cursor, Cline, and any other MCP-aware client can use it.
CPP's engine is a public MCP server at mcp.criticalpathpartners.ca — thirteen forensic tools, no signup required.
Anthropic's framework for orchestrating multi-step AI workflows — the layer where Claude decides which tool to call next, reads what came back, and sequences the analysis instead of running one shot.
CPP uses it for the deeper deliverables that chain four or five tools — windows analysis into collapsed-as-built validation into Monte Carlo into claim packaging.
Each Skill is a self-contained capability — code, methodology references, trigger documentation. Claude loads only the ones the task requires. The thirteen marked MCP are also exposed through the hosted server; the rest produce DOCX, XLSX, and PDF deliverables and run inside engagement work.
Canonical Primavera P6 XER file parser and generator. The foundation every other schedule skill consumes — parses the table-by-table structure, normalizes calendars, resolves predecessor relationships. Also writes XER for downstream tools.
Schedule Health Dashboard from baseline + current XERs. DCMA-14 metrics, baseline-vs-current slip register, charts, embedded reproducibility manifest — single self-contained HTML.
Critical path validation, logic health assessment, optimization recommendations. Detects false criticality from constraints, open ends, near-critical paths — the diagnostic for "is my critical path actually right."
Logic-trace driver chain explorer. Answers "why is this activity critical" by walking driving predecessors back to project start, and "what does this drive" forward to finish. MCPM-aware; cites AACE 24R-03 §4 for constraint-driven artificial criticality.
Stand-alone forensic delay report — windows analysis (AACE MIP 3.3, contemporaneous as-is). HTML dashboard, DOCX executive report, TXT narrative, per-window and per-activity CSVs. Contractor perspective against owner.
Collapsed As-Built / But-For analysis (AACE MIP 3.8 — modeled, subtractive, single simulation). Independent dual-method validation against windows analysis. The but-for opposing counsel always demands.
Prospective fragnet insertion (AACE RP 52R-06 / MIP 3.7). Models a delay event into a CPM network, re-runs the forward pass, quantifies completion impact. Forward-looking EOT analysis for RFIs and change orders.
Forensic workbench. One folder of mixed evidence (XER chain plus MSG / PDF / DOCX / XLSX correspondence) becomes evidence ledger, chain-diff manipulation log, rolling baseline, trust score, NER-extracted recurring actors, and a slip-to-evidence cross-reference.
Full-pipeline owner-submission EOT umbrella. Chains forensic-delay-analysis plus claims-preparation, adds component completion schedule and owner-submission QC. Produces the entire submission deliverable set.
Construction claims preparation and packaging. Cover letters, per-event exhibits, forensic workbook, mitigation report, supporting documents index — turns forensic findings into a formal submission document.
Defense analysis from the receiving party's perspective. Tests an opposing claim against ten defense theories — concurrent delay, pacing, no notice, failure to mitigate, causation gap, float ownership, scope-change disguise, self-caused acceleration, LD cap, no CP impact.
Monte Carlo schedule risk. P10 / P50 / P80 / P90 completion forecast, sensitivity tornado, mitigation-vs-baseline scenario comparison. Output JSON feeds claims-preparation directly. AACE 122R-22 QRAMM badge included.
Per-resource time-phased usage at day, week, or month granularity. Peak headcount, average crew size, overallocation periods, trade-stacking detection, aggregate workforce histogram. The analytical foundation for lost-productivity damages.
Multi-project rollup. Aggregates 5–50 monthly progress reports into one self-contained HTML — RAG status, SPI / CPI rollup, finish-drift heatmap, top-five at-risk projects, aggregate critical-path roll-up.
Monthly progress report — HTML dashboard plus DOCX executive report. Four-axis snapshot: variance against baseline, progress this period, upcoming 30-day work, current overall status. Consumes current XER plus optional baseline and prior-period XER.
Standardized 3-week lookahead workbook from a P6 XER. Four tabs — overview plus three week-by-week breakouts. Δ-vs-last-week variance and DROPPED appendix when a prior workbook is supplied. Format is locked.
Static PDF reports. Email attachments. Workbooks shipped through file-transfer portals. Zero programmatic access. The deliverable is a binder. That's the industry norm and it has not meaningfully shifted in a decade.
When AI assistants become the default research interface for construction lawyers, project managers, and claims consultants — already happening — the firms with MCP servers and Claude Skills get found first. A construction lawyer in Toronto asking Claude "run a DCMA-14 health check on this XER and tell me which contractor's schedule looks defensible" is going to call the tool that's actually wired in.
Twenty-five years of forensic methodology plus the Claude-native stack is a moat that takes a real practitioner months to replicate. Not because the AI is hard — the AI is the easy part. Because writing the methodology in a way that a tool can execute correctly, defensibly, and without fabricating case citations requires having actually written the claim, sat through the cross-examination, and watched the arbitrator's pen move. The skills carry the disclosures, the heuristic gates, the conservation-rule attribution math, and the audit-trail discipline that comes from twenty-five years of doing the work.
Wire CPP's tools into your own Claude session in 30 seconds. JSON snippet, copy-paste, done. Same thirteen tools whether you use claude.ai in the browser or Claude Desktop.
Try the engine yourself. Browse the full suite. Or skip both and put a real claim in front of someone who's done the work.