Marfin
    CatalogPacks
    GitHub
    INTERMEDIATERESEARCH07 de mai. de 2026

    Stack Cost Compare

    Compares 2-4 stacks for YOUR workload with projected cost, hidden gotchas (NAT Gateway, egress, CPU limits), trade-offs per dimension and a recommendation with explicit confidence + a trigger to re-evaluate.

    View on GitHub
    WORKS IN
    CLAUDEany setup

    Pure-LLM skill. No MCP, no paid tool required. Works on Claude.ai web, desktop and Claude Code.

    WORKS BEST WITH
    • Web Search (built-in)optional

      providers change pricing constantly — without web search the skill quotes prices from 6+ months ago (wrong). Does a 1-pass search of the official pricing pages before generating

    • Stripe MCPoptional

      if you already pay for hosting through Stripe, pulls the current bill to anchor the comparison to a real number

    TAGS
    founderstackcostinfracomparison

    What this skill does

    Compares stacks (Vercel vs Cloudflare vs AWS vs Hetzner, Supabase vs Neon vs self-hosting, etc) with cost projected for your volume and the real trade-offs (lock-in, latency, DX). No fanboyism, no marketing.

    What you get

    • Comparison table: cost at your volume + how it scales
    • Honest trade-offs (DX, lock-in, regional latency, learning curve)
    • A recommendation with a confidence level (high/medium/low)
    • A list of 'gotchas' per option (e.g. AWS NAT Gateway is expensive)
    • A migration plan in case you want to switch later

    When to use

    • Before picking a stack for a new project
    • When the current bill hurts and you're considering migrating
    • To answer the investor's 'why do you use X?'

    How it works

    1. You give it: the workload (HTTP req/min, DB query/sec, storage GB, regions)
    2. The skill computes cost across 3-4 typical stacks at your volume
    3. Cross-references known 'gotchas' (e.g. Vercel function 10s timeout)
    4. Generates a trade-off table by dimension
    5. Makes a recommendation with an honest confidence level
    6. Suggests a re-evaluation point (which volume triggers a switch)

    Output example

    Stack Cost Compare — Marfin workload

Inputs
HTTP: ~50k req/day (peak 200/min)
DB: 30 query/sec avg, 10GB storage, 5-year retention
Functions: 3-5s avg, some 30s
Region: BR + EU users

                   Vercel    Cloudflare   AWS       Hetzner
Hosting             $20       $5          $35       $5
DB (Postgres)       —         Hyperdrive  RDS $40   self R$ 30
                              + Neon $20             + backup $5
Functions           included  included    Lambda $5  N/A
Egress              $0        $0          $9        free
TOTAL/mo            $40       $30         $89       $10

Trade-offs by dimension
DX                  Vercel    Cloudflare   AWS       Hetzner
                    ★★★★★    ★★★★         ★★         ★
Lock-in             med       low          high      none
Latency BR          ~150ms    ~50ms        ~200ms    depends
Time to debug       easy      medium      hard      manual

Recommendation: Vercel (high confidence)
DX and iteration speed are worth the extra $30. For a solo founder,
1h spent debugging AWS already pays for 10 months of Vercel.

Re-evaluate when:
- Bill > $300/mo (10x the volume)
- BR latency becomes the bottleneck (consider Cloudflare)

Known gotchas
- Vercel: 300s function timeout on Pro (are you on Pro?)
- AWS: NAT Gateway $32/mo fixed, shows up as a surprise
- Cloudflare Workers: 10ms CPU limit on free
- Hetzner: backup is DIY, data center maintenance downtime

Migration plan (Vercel → Cloudflare)
1. DB first (Neon → R2/D1) — 1 week
2. Functions (Vercel → Workers) — 2 weeks
3. Front end (Vercel → Pages) — 3 days

    Pre-reqs

    None required. Web search helps pull up-to-date pricing (providers change prices often).

    Build.

    Work.

    Dream.

    Create.