TwinTrack is TwinStack's modular business platform, and Claude Code is its main developer. Claude writes and maintains the code, keeps documentation and tests in step with every change, and fixes failed deploys automatically. Claude can also work with TwinTrack data through an MCP server.
At a glance
| Type | Internal product |
|---|---|
| Claude products | Claude Code (CLI and on the web), Claude Code hooks and skills, Claude Code GitHub Action, MCP connectors |
| Tech stack | Node 22, Express, TypeScript, MongoDB, Next.js, Vitest |
| Deployment | Automatic test and deploy to demo.twintrack.work on every push to main |
The challenge
TwinTrack has one backend serving many modules, each with its own frontend, and fine-grained permissions (module access, scopes, record sharing and field-level security). On a platform like this, every backend change must also update route manifests, API collections, test scenarios, docs and handoff packs. Keeping all of that in step by hand is slow, and easy to get wrong.
The solution
TwinStack made Claude Code the platform's main developer, and gave it the rules, tools and guardrails to work safely.
| Where | What Claude does |
|---|---|
| CLAUDE.md files | Give every Claude Code session the architecture, the route checklist, coding standards and commit rules |
| twintrack-sync skill | On any backend change, keeps manifests, Postman collections, test scenarios, API docs, the progress page and module handoff packs in step |
| Claude Code hooks | Remind Claude of the skill at session start, flag the first backend edit, and block finishing if the backend changed but its documentation didn't |
| Claude Code on the web | Runs typecheck, lint and tests, pushes to main, then watches the deploy run and fixes it until it is green |
| Claude Code GitHub Action | If the demo deploy fails after a person's push, reads the failed log, fixes the code, re-checks, pushes and reruns. Bot pushes are not retried, which prevents loops |
| Module starter and handoff packs | Each pack carries its own CLAUDE.md, data model and instructions, so different people building module frontends with Claude Code get consistent results |
| Claude artifacts | Keep the progress page and API scenarios page current after each change |
The change flow:
- A developer asks Claude for a change, and Claude edits the backend.
- A hook triggers the twintrack-sync skill, which updates the manifest, Postman collection, test scenarios, docs and packs.
- Checks run, and Claude pushes to main.
- CI tests the code and deploys it to the demo site.
- If the deploy fails, Claude in CI fixes it.
Claude as a client, through MCP
TwinTrack also exposes its own MCP server. Tools include checking your permissions, finding colleagues and listing clients, and each module can add its own. Claude Code connects with a personal API token, and claude.ai connectors use OAuth. Tools run with the caller's own profile, so Claude sees only what that person is allowed to see.
Results
- Code, documentation, API collections and test scenarios stay in step on every change, enforced by hooks rather than memory.
- Failed deploys are fixed automatically, with a safeguard against loops.
- Automated tests (Vitest with an in-memory MongoDB) check that routes match their manifests, that Postman coverage is complete and that the test-page scenarios still pass.
- Handoff packs let different team members build new modules with Claude Code and get consistent results.
- People can query platform data from Claude through MCP, within their own permissions.
Want Claude working inside your business?
TwinStack builds Claude agents, skills and integrations that fit how your team already works, with people kept in the loop where judgement matters.
Talk to Us