life. scored. MCP server

life. scored.

You are already a number.

A credit score, an actuarial row, a callback probability — this app rebuilds those numbers in the open: every rule cited or flagged as a guess, every weight visible and editable. Transparency, not judgment.

check it out - lifescored.com

Stack

deterministic local fallback (the app is fully functional with no API key at all)

• Svelte 5 + SvelteKit, prerendered static pages, deployed to Cloudflare Workers
• One dynamic endpoint: POST /api/narrative — KV-cached Gemini narratives with a

• All user data stays in the browser (localStorage + URL-fragment share links)
• Traffic is measured with Cloudflare Web Analytics — cookieless and aggregate; it cannot see your inputs, which never leave your device anyway.

Develop

npm install npm run dev # app at localhost:5173 (narrative falls back to local composer) npm test # vitest suite: rulebook invariants, engine, codec, worker handler npm run check # svelte-check npm run types # regenerate Cloudflare bindings after changing wrangler.jsonc

Narrative cache misses reserve one call atomically in the NarrativeBudget SQLite Durable Object: at most 200 per UTC day globally and 10 per IPv4 address or IPv6 /64 network per day. IPv4-mapped IPv6 shares the IPv4 allowance. People on the same network may share an allowance; cache hits and the local composer remain available. Failed upstream calls consume a reservation; failed reservations use the local fallback. Cache hits do not consume quota. The coordinator stores only a daily address hash, and removes previous-day entries on the next day's first reservation. It never sees profile inputs or narrative text. KV is used only for the narrative cache.

npm run build adds the Durable Object export to the adapter-generated Worker; deploy that build through the existing workflow. Wrangler creates the coordinator through the versioned migration in wrangler.jsonc. Run npm run types after binding changes; its helper resolves the class from source so checks work on a clean clone.

Deploy (Cloudflare free tier)

Deploys run automatically from GitHub Actions on every push to main (.github/workflows/deploy.yml): the workflow runs the full test suite + typecheck, and only then builds and deploys. Pull requests run the same gates and a Worker dry run without deploying.

One-time setup:

Secrets are kept in hush: stored once in your OS keychain, then piped straight into the consumer — they never get pasted into a terminal, echoed, or printed. This project's hush items use a lifescored- prefix so they group together in a keychain search. (No hush? Drop the hush pipe … -- prefix and run the bare npx wrangler / gh command; hush just wraps it.) Local dev needs no secrets at all — the narrative falls back to a local composer.

1. Create the KV namespace and paste its id into wrangler.jsonc:

npx wrangler kv namespace create NARRATIVE_KV

omit to run AI-free with the local narrative fallback):

1. Set the Gemini key as a Worker secret (persists across deploys; optional —

hush set lifescored-gemini-key # paste it once, hidden dialog hush pipe lifescored-gemini-key -- npx wrangler secret put GEMINIAPIKEY

(Settings → Secrets and variables → Actions):

1. Give GitHub Actions the deploy credentials

hush set lifescored-cloudflare-api-token # "Edit Cloudflare Workers" token hush pipe lifescored-cloudflare-api-token -- gh secret set CLOUDFLAREAPITOKEN gh secret set CLOUDFLAREACCOUNTID --body "" # account id isn't secret

Shortened. The full README is on GitHub.