Back to News

Data Sandbox grows up: 38 personas, Insurance Data Sharing GA, an ATM directory, and an MCP server

When the Open Finance Data Sandbox launched in April, it covered Bank Data Sharing with 10 synthetic personas. Six weeks later, it has become a much bigger thing — covering all three UAE Open Finance data domains with nearly four times the persona library.

What’s New

38 synthetic personas (up from 10)

The library now spans 21 banking-only personas — retail archetypes plus 8 SME profiles (cash-heavy retail, F&B multi-outlet, e-commerce marketplace, free-zone SaaS, construction sub-contractor, healthcare clinic, RAK Emirati-owned trading, general trading) and a corporate treasury — 9 insurance-only personas, and 8 multi-domain personas that emit both banking and insurance payloads for the same customer. The banking view renders 29 personas; the insurance view renders 17. A 39th manifest anchors the new ATM Locator domain with a public directory of LFI-operated cash machines.

Insurance Data Sharing, GA across all 7 lines

Motor, home, health, life, travel, renters, and employment — 30 endpoints (policies, policy detail, payment details, and quote per line, plus 2 Insurance Consents endpoints), all generated and validated against the v2.1 OpenAPI schema, alongside the original 12 Account Information endpoints.

Multi-bank, multi-insurer personas

The flagship retail multi-banker’s realistic footprint spans four banks (salary primary, everyday card, digital sidekick, mortgage lender) and three insurance carriers — with role-keyed bundles and mod-97-valid IBANs. No single bank sees the full debt-service picture; only Open Finance aggregation across all four does. It’s exactly the persona shape consent-flow and underwriting demos are built for, and most SME personas declare multi-LFI footprints too.

An MCP server for AI agents

@openfinance-os/sandbox-mcp exposes the whole corpus as MCP tools, resources, and prompts, so an AI agent can act as a personal-finance assistant over a synthetic customer — answering balance, spend, obligation, and coverage questions from deterministic v2.1 JSON. Use the hosted endpoint (https://data-sandbox.fly.dev/mcp as a custom connector in Claude.ai — anonymous, no API key) or run it locally with npx -y @openfinance-os/sandbox-mcp. A custom persona builder lets you compose your own profiles beyond the curated set.

Pinnable fixture packages

The synthetic dataset ships as versioned packages your CI can pin: @openfinance-os/sandbox-fixtures on npm and openfinance-os-sandbox-fixtures on PyPI — v2.1-shaped JSON envelopes across the persona × LFI × endpoint matrix, with the same determinism guarantees as the browser explorer.

TPP integration guide

A worked integration guide covers four plug points for wiring sandbox personas into a TPP demo journey: iframe embed, npm loadJourney(), PyPI load_journey(), and raw HTTPS fixtures.

Unchanged — and still the point

Everything remains zero-setup, client-side, and deterministic: the same persona, LFI profile, and seed produce byte-identical payloads on any machine, any day. The three populate-rate profiles (Rich, Median, Sparse), the side-by-side LFI comparison view, the underwriting scenario panel, full-spec search, exports, and embed mode all carry over — now across three data domains.

Try It

Visit data-sandbox.openfinance-os.org — no registration, no credentials, no install — or read the full overview on Commons.

The sandbox is open source on GitHub under the MIT license; synthetic data and fixtures are released under CC0. Contributions and issue reports are welcome.

All News