Common use cases
Four scenarios that cover most of how teams actually use Tracker. Pick the one closest to your situation; the steps assume you've done Getting started already.
1. Get EU AI Act ready before your high-risk system goes live
Most common use case for Tracker. You're shipping an AI product into the EU, you've concluded it's high-risk under Article 6, and you need a defensible compliance posture before deployment (Article 6 hits August 2026; some prohibitions are already in force).
- Create one project per AI system in scope. Don't lump multiple systems together — each system gets its own conformity assessment.
- Import the EU AI Act framework pack. Articles 9 (risk management), 10 (data governance), 11 (technical documentation), 12 (logging), 13 (transparency), 14 (human oversight), 15 (accuracy / robustness / cybersecurity), 16 (provider obligations), 17 (QMS), 26 (deployer obligations) and 50 (transparency for certain systems) all land as ready-to-track controls.
- Walk top-to-bottom. For each requirement: capture what you already have as evidence (PR URL, doc link, audit-log export); log a gap where you don't; spin gaps into tasks with a target date.
- Layer the NIST AI RMF and OWASP LLM Top 10packs alongside if your system is generative — they cover the operational and security territory the AI Act gestures at but doesn't prescribe.
- Use the project dashboard to track Article-by-Article completion. A fresh import of the EU AI Act pack is ~30 requirements; a focused team closes the major gaps in 4–8 weeks.
What this gets you: a structured record of how each Article is satisfied, with evidence linked to actual changes, and a gap register for everything outstanding. That record is what you hand to a notified body or your legal counsel when conformity assessment time comes.
2. Stack ISO 42001 alongside SOC 2 for enterprise procurement
You're selling AI to enterprise. Their procurement asks for SOC 2 (security posture) and increasingly for ISO 42001 (AI management system). Both need a paper trail; you don't want two parallel tracking systems.
- Import the ISO 42001 pack. Clauses 4–10 (the management-system shell) plus the Annex A controls land directly as project requirements.
- Use a separate platform (Vanta, Drata, Secureframe — whatever procurement recognises) for SOC 2 itself. Tracker complements rather than replaces the SOC-2 layer.
- Where a control overlaps — risk management, access control, data governance, incident response — treat the SOC 2 evidence as also satisfying the ISO 42001 requirement. Add the SOC 2 audit reference as evidence on the ISO 42001 requirement; the audit log captures the cross-link.
- Layer ISO/IEC 23894 (AI risk management process — the operational counterpart to 42001) if you need depth on the risk-management clauses.
What this gets you: one compliance surface that satisfies both audits, with explicit cross-references where a single piece of evidence covers multiple frameworks.
3. Put compliance in the AI agent's context with Claude Code
If your team uses Claude Code (or any MCP-aware agent), evidence logging becomes part of the work itself — the agent sees which requirements your edits affect and can log evidence inline, without anyone having to come back to it later.
- Wire the MCP per Claude Code + MCP. Mint a token, set
.env, runclaude mcp add tracker. - Drop
docs/claude/skills/tracker-consult.md(in this repo) into your consuming project's.claude/skills/. It teaches the model the consult-before-edit ritual. - When the agent starts a session, it auto-reads three resources: your active requirements, your open gaps, and the path-map (which file globs are load-bearing for which requirements).
- Before any non-trivial edit, the agent calls
check_changewith the paths it's about to touch. The response tells it which requirements are affected, surfaces any open critical gaps, and suggests evidence to capture. - After the change ships, the agent (or you) calls
add_evidencewith the PR URL and the touched paths infilePaths. Futurecheck_changecalls now know those paths map to that requirement — the path map sharpens with each evidence-tagged change.
What this gets you: compliance evidence accrues as a byproduct of normal development, not as catch-up work the week before an audit. The audit log on your org records every consultation (mcp.context.read, mcp.check_change.called) so you can prove the agent was informed before each change.
4. Prepare an audit-ready evidence package for a customer or auditor
A customer asks “show me your AI governance posture” or an auditor schedules a review. You need to produce a coherent package fast, without scrambling.
- Pick the project (or projects) in scope. Make sure each requirement has either evidence or a gap with a target date — incomplete records read worse than honest gaps.
- Use the CSV export on requirements / gaps / tasks pages to get a snapshot the auditor can ingest into their own tooling. Each row carries status, owner, dates, and the evidence-count column.
- Walk the auditor through the project dashboard live. Framework completion percentages, gap severity breakdown, and the recent activity timeline are all visible at a glance.
- Show the audit log (Organizations & teams covers permissions). Append-only at the database layer with IP/user-agent on each event — the kind of immutability auditors ask for and rarely get from spreadsheet-based systems.
- For each gap they probe, surface the linked tasks and target dates. A gap with a named owner and a reasonable plan beats a requirement marked “met” with no evidence.
What this gets you:a credible narrative of where you are, where you're going, and how you got there — backed by an immutable record. PDF compliance-package export is queued for a future release; until then the CSV exports plus a screen-share cover the same ground.
Next
- Wire the MCP for the AI-agent compliance flow.
- Self-host Tracker if you need it in your own infrastructure.
- Invite collaborators and assign roles.