SpendGuard AI

A production-shaped MCP portfolio application. SpendGuard AI helps a FinOps team investigate cloud-spend anomalies, draft a savings action, obtain independent approval, execute it through a safe simulated provider, survive retries and restarts, and produce an authoritative receipt plus operational evidence.

The default provider is simulated and deterministic. This public portfolio never spends money, mutates a real cloud account, exposes customer data, or requires a real cloud credential.

Live

• GitHub: https://github.com/ketankshukla/spendguard-mcp
• Production: https://spendguard-mcp-web.vercel.app/

Monorepo Layout

apps/web           Next.js App Router product (marketing + product UI, MCP transport, AI host)
services/risk-engine  Independently deployable Python MCP server (proposal scoring, simulation)
packages/           domain, contracts, db, auth, policy, mcp-server, mcp-host, registry, config, ...
docs/               implementation-plan.md, worklog.md, ADRs, runbooks

Local Development

pnpm install
pnpm turbo lint typecheck test build
pnpm --filter web dev

Requires Node.js >= 20, pnpm (via Corepack or npm install -g pnpm), and — for the Python risk service — Python 3.14 and uv.

Status

See docs/implementation-plan.md for phase-by-phase status and docs/worklog.md for a running log of decisions, commands, and results.