PILLAR_R02 LIVE
Recipes that actually saved time.
Chained workflows. Multiple agents and MCPs working together. Every recipe has a measured before/after — no vibes, just numbers.
5 recipes ready
Automate YouTube SEO across 30 videos
Three MCPs chained: rewrite titles in Sush's voice, push updates, verify cache.
Before 2 hrs / month
After 8 min
Cost / run $0.40 / run
playwright-mcp youtube-mcp cloudflare-mcp
Production Open recipe →
PR triage with 3 MCPs
Auto-labels, suggests reviewers, summarises diffs into Slack channel.
Before 45 min / day
After 4 min
Cost / run $0.18 / run
github-mcp slack-mcp claude-api
Production Open recipe →
Toolkit data freshness check
Cron-driven audit of TOML data files; flags stale entries before users hit them.
Before 4 hrs / month
After 0 min
Cost / run $0.05 / week
github-mcp filesystem-mcp claude-api
Production Open recipe →
Blog SEO audit at scale
Search-MCP scans live SERPs, returns gaps + suggested rewrites for each post.
Before 1 hr / post
After 5 min / post
Cost / run $0.12 / post
web-search-mcp cloudflare-mcp claude-api
Production Open recipe →
MCP health monitoring (the meta one)
The cron pipeline that powers this site's own health board. Eat your own dog food.
Before manual checks
After every 6h
Cost / run $0 (Actions)
github-mcp filesystem-mcp
Production Open recipe →
WHAT MAKES A RECIPE
The format
- Outcome-focused title — "Automate X in Y minutes", not "Using Tool Z"
- The components — every agent, MCP, and API listed
- Architecture diagram — visual of how they chain
- Before/after table — measured numbers, no estimates
- Step-by-step build — actually run-able
- The catch — every workflow has one
- Cost — tokens, API, real money per run
- Sush's take — would I run this again?
If a recipe doesn't have measurements, it's a draft. We don't ship vibes.