The Growth Stack is 6 MCPs (Postgres, Slack, Perplexity, Notion, Firecrawl, GitHub) that cover research → experiment → analysis for a growth team. Ideas come from Perplexity + Firecrawl; implementation runs through GitHub + Notion; results get analyzed in Postgres and announced in Slack. Shipping 2–3x more experiments per month is typical.
Analytics, research, notes, and copy — the growth team's AI backbone
$ mcpizy install postgres slack perplexity notion firecrawl githubOne command installs and configures all 6 MCPs for Claude Code, Cursor, Windsurf, or any MCP-compatible client.
Growth teams are research-driven and experiment-heavy. The bottlenecks are: finding what to test (research), running the test (implementation), and analysing results (analytics). This stack covers all three. Perplexity + Firecrawl drive research; GitHub + Notion manage implementation and experiment docs; Postgres + Slack analyze and broadcast results.
The multiplier: a growth team with this stack runs 2–3x more experiments per month, because the ideation-to-implementation friction collapses. Claude drafts the experiment doc, writes the landing page variant, opens the PR, sets up the PostHog event, and ships the Slack update — all in one session.
Replaces ~$500/mo of growth tooling (Visualping, Klue starter, research VA hours) for a 3-person growth team. Bigger win: experiment velocity doubles, which compounds into faster CAC reduction.
PostHog MCP exists (community). Amplitude MCP is in early beta. For now, the Postgres MCP in this stack assumes you export product analytics to Postgres via Fivetran/RudderStack. Direct MCPs will add cleaner event-level access when they mature.
Yes — given a design spec (Figma/Notion) and the existing component code (GitHub MCP), Claude writes the variant, wires PostHog events, and opens the PR. Most growth engineers cut variant-creation time by 70%.
Both have community MCPs in early stages. For production-ready growth sites, keep landing pages in GitHub (Next.js/Astro) for cleanest MCP integration. Webflow MCP works but is limited today.
Yes — the stack is domain-agnostic. B2C teams often add a creative/ads MCP (Meta Ads MCP, TikTok Ads MCP) for paid-heavy work. The core 6 stay the same.
Optimizely gives you the runtime traffic-splitting infra. This stack gives you the ideation + analysis infra. They complement each other — use Optimizely for sophisticated splits, MCPs for the 'Claude writes the variant' step.