MY:TIME — Competitive Intelligence Platform

Ingestion pipeline + remote MCP server over a Supabase Postgres single source of truth. Crawls competitor e-commerce sites and infers demand from inventory depletion, tracks social performance, and exposes everything to Claude, Power BI, and the team through one MCP endpoint.

Honesty note: depletion-derived sales are estimates, labeled as such in all outputs. Competitor social = public metrics only. MY:TIME's own data comes from official APIs and the Adform product feed, never guessed.

Architecture

SOURCES      competitor e-commerce sites + social platforms
COLLECTION   Apify + FireCrawl + official social APIs + MY:TIME XML feed
INGESTION    Node/TS pipeline — route, normalize, dedupe, date-stamp (modular per source)
STORAGE      Supabase Postgres (Frankfurt) — single source of truth
MCP SERVER   remote, Streamable HTTP, https://mcp.mytimeprime.mk  ← the hub
CONSUMERS    Claude · Power BI / dashboards · internal team

Related MCP server: Metrx MCP Server

Workspace layout

Stack

Node 20+ · TypeScript (ESM, NodeNext) · pnpm workspaces · Drizzle ORM · Biome (lint + format) · pino logging · zod validation · Supabase Postgres.

Getting started

corepack enable && pnpm install      # pnpm 11.x
cp .env.example .env                 # fill in credentials (gitignored)
pnpm build                           # build all packages
pnpm validate:targets                # validate config/targets.json
pnpm lint                            # Biome

Every runtime variable is documented in .env.example; missing required vars fail fast naming the variable.

Adding a new data source (modularity contract)

A new source is new module + new table + new MCP tool, with no edits to existing sources. See ingestion/src/sources/README.md.

Build phases

• Phase 0 — Scaffold (current): monorepo, tooling, targets.json schema, env.

• Phase 1 — Site profiling & crawler selection → crawler-plan.md

• Phase 2 — Database schema (full time-series model + ER diagram)

• Phase 3 — Ingestion pipeline — A–G done: idempotent writer; all 9 web sites live (WC Store API, Adform feed, sitemap+JSON-LD, Zia API, FireCrawl-map+OG, Magento listing, nopCommerce); competitor social (Apify IG/FB/TikTok); own-brand social (MY:TIME IG/FB via the official Meta Graph API, meta-own-brand collector, see docs/step-f-own-brand-social-setup.md); daily scheduler (systemd timer + runbook, see deploy/).

• Phase 4 — Demand inference + MCP server core — depletion engine (exact/assumed qty_basis) + Streamable HTTP MCP server with 4 tools (get_inventory_velocity, compare_market_share, social_benchmark, price_assortment); see mcp-server/.

• Phase 5 — OAuth 2.1 + two-layer authorization — Google upstream via a custom MCP OAuthServerProvider (PKCE/S256, DCR, Protected Resource Metadata, 401+WWW-Authenticate); Layer 1 signed-Google-ID-token domain gate (@mytime.mk), Layer 2 authorized_users whitelist/role, per-tool role enforcement (admin/analyst/viewer). Needs your Google OAuth client (see docs/phase-5-google-oauth-setup.md) for the end-to-end browser flow.

• Phase 6 — Deploy to Hetzner (Caddy, TLS, daily ingestion)