Self-Host n8n vs Zapier for Indian MVPs in 2026
By Rohit Raj — Founding Engineer, 6 years shipping production MVPs · LinkedIn
The short answer: if your MVP automates more than ~2,000 operations a month and you can run one Docker container, self-host n8n. For an Indian MVP doing 5K-50K operations/month, self-hosted n8n on a $7 Hetzner VPS costs ₹600/month all-in; Zapier Professional at the same volume costs ₹6,100/month — a 10x premium for the same JSON-pushing. Below 2,000 ops/month Zapier wins on time saved; above it, n8n's ₹5,500/month delta funds half a junior contractor.
The cost gap is structural, not promotional. Zapier charges by task count because its margin is engineering operating a managed SaaS — observability, multi-region failover, vendor SOC 2, integration certs. n8n is open-source MIT-licensed software you run on a single VPS. The marginal cost per workflow run on your own box is electricity and bandwidth, both negligible at MVP scale.
The structural reason: Zapier's pricing model assumes you cannot or will not run infrastructure. The moment you can run a $7/month Hetzner CX22 VM with Docker, the SaaS premium evaporates. India-side, a Hetzner CX22 in the EU costs ₹580/month including VAT after the 2026 INR rate, plus ₹0 in egress to Indian endpoints under their generous free tier. That's the entire bill for unlimited tasks.
The real cost math at 50K operations a month
Yesterday I finished migrating my own social-media dispatcher to self-hosted n8n. Five platforms (X, Facebook Page, Reddit, LinkedIn, Bluesky), one webhook trigger, Claude drafts text via the Anthropic API, and a refinement loop fires when a human edits. The workflow runs roughly 35-50 times a day across drafting + cross-posting, peaking around 1,500 operations/month per platform = 7,500 total. Plus a tail of error-retry runs that doubles real ops count to ~15K.
I ran the same logic on Zapier Free for two weeks first to baseline cost. Here is the actual rupee delta.
| Metric | Zapier Professional | Self-hosted n8n |
|---|---|---|
| Plan tier needed for 50K ops | Professional ($73/mo) | Free (MIT licensed) |
| Compute cost | Included | Hetzner CX22 ₹580/mo |
| Database | Included | Postgres on same box ₹0 |
| Bandwidth | Included | Hetzner 20TB free ₹0 |
| Backup | Included | Hetzner snapshot ₹95/mo |
| Currency conversion | 3% FX hit on USD card | 0% (INR direct debit) |
| Total monthly cost | ₹6,156 | ₹675 |
| Annualized | ₹73,872 | ₹8,100 |
| Delta | — | ₹65,772/year saved |
That ₹65K/year is roughly three months of an OpenAI Tier-1 API budget for an MVP, or one good MacBook Air refresh, or a junior contractor for two and a half weeks. It is not nothing.
Important caveat for Free-tier Zapier users: Zapier Free only allows 100 tasks/month and only single-step Zaps in 2026. Anything multi-platform like the dispatcher above forces the Starter plan ($30/mo) minimum, and the moment you cross 750 tasks/month you hit Professional. The free tier is a marketing funnel, not a real plan.
Operational reality: Docker compose is a 30-minute setup
The Zapier pitch is that self-hosting is hard. In 2026 it is not. The full n8n deployment I ran yesterday on a fresh Hetzner CX22 took 28 minutes from VM provision to webhook live. Here is the exact docker-compose.yml:
The ops surface that follows: one Caddy reverse proxy for HTTPS (Caddy auto-renews Let's Encrypt certs every 60 days, zero config). One nightly cron for Postgres dump to S3-compatible cold storage (Wasabi at ₹500/year for the first 1TB). One UptimeRobot HTTP check (free tier, 5-minute interval). Total ongoing maintenance: about 10 minutes a month rotating credentials and inspecting logs.
The trap most "self-host is hard" articles miss: ops is hard if you DIY observability + auth + auto-scaling. For a single-tenant MVP webhook box, you do none of that. The Docker container is the entire stack. If it crashes, docker compose up -d restarts it. If you lose the box, the snapshot restores in 90 seconds.
When self-host hurts: the hidden tax nobody mentions
I do not want to oversell this. Self-hosting has real friction Zapier handles for you.
Integration coverage gap. Zapier has 7,500+ pre-built app integrations in 2026 including obscure SaaS like FreshBooks Classic and Manychat legacy webhooks. n8n has roughly 1,000 first-party nodes. For the long tail, you fall back to the generic HTTP node and hand-roll OAuth flows. That LinkedIn integration on n8n took me 3 hours because I had to manage the OAuth refresh token myself. On Zapier it would have been 4 clicks.
Error replay and observability. Zapier shows a polished task history with replay + retry buttons. n8n has execution history but the UI is rougher, and notifying yourself about a failed run requires wiring up either the Error Workflow feature (one extra workflow) or piping logs through Sentry/Better Stack. I spent another 90 minutes on this for the dispatcher.
Webhook routing during scale spikes. Zapier transparently buffers spikes. n8n on a CX22 will start dropping webhook requests above roughly 50 requests/second sustained. For a B2C webhook receiver during a viral moment, this matters. The fix is queue mode (n8n + Redis + worker pool), which adds an evening of setup work and roughly ₹200/month in extra compute.
Concrete real-world hit: one of my workflows reposts to LinkedIn. LinkedIn rotates their OAuth scopes about every 8 months. Each rotation breaks my n8n LinkedIn node and I spend 45 minutes re-authenticating. Zapier abstracts these breakages — their team patches the connector globally. That is the ₹65K/year you are buying back.
Side-by-side comparison
| Dimension | Zapier Professional | Self-hosted n8n |
|---|---|---|
| Pricing model | Per-task ($73 for 50K tasks) | Flat infra (₹600/mo all-in) |
| Setup time | 5 minutes (web UI) | 28 minutes (Docker + DNS) |
| Pre-built integrations | 7,500+ | ~1,000 first-party |
| Custom HTTP / OAuth | Available but slower UI | Native, programmable |
| Webhook throughput | Buffered transparently | ~50 req/s before queue mode |
| Error replay UI | Polished | Functional, less polished |
| Conditional logic | Filter step + Paths | Code node (JS or Python) |
| LLM nodes (Claude/GPT/Gemini) | Native, 2026 update | Native, MCP-server compatible |
| Logs retention | 30 days | Forever (your Postgres) |
| RBAC for teams | Yes (Professional+) | Single-user / Basic-auth |
| SOC 2 / HIPAA | SOC 2 Type II, HIPAA BAA | DIY (you are the auditor) |
| Vendor lock-in | High (workflow export is JSON-ish) | Zero (workflows are JSON files) |
| Currency exposure | USD billed, 3% FX cost | INR (Hetzner Indian invoicing) |
| Compliance for Indian data | Data leaves India (US/EU) | Pin Hetzner region; can stay EU or use Indian VPS like E2E Networks |
The AI-node gap closed in 2026. Both platforms have first-class Anthropic, OpenAI, and Gemini nodes now. n8n shipped MCP server support in March 2026 which means agentic workflows. Zapier has Claude tool-use integration but it's coupled to their Tables feature. If you are building agentic flows that touch your own MCP servers, n8n is strictly better.
When Zapier still wins (and you should pay the premium)
Stay on Zapier when any of these are true. I have shipped MVPs in all four conditions where switching to self-host would have been malpractice.
1. You are pre-PMF and time is the binding constraint. If a one-hour ops detour means missing a customer demo, do not self-host. The ₹65K/year is a future-you problem. Pay the SaaS tax now.
2. You need SOC 2 attestation for an enterprise deal. Zapier's audit reports satisfy procurement. Self-hosted n8n requires you to either run your own pentest + SOC 2 process (₹3-5L/year), or buy n8n Cloud Enterprise (which is just managed n8n and costs $50/user/month — the worst-of-both option). For Indian MVPs selling to US enterprise, Zapier is the cheaper compliance path until you have ARR to fund your own audit.
3. You will not maintain a Docker container. Even 10 minutes/month is a commitment. If your team is non-technical or you are a solo founder spending all your time on customer development, the box will rot. Zapier never rots.
4. Your integration count exceeds n8n's coverage. If you genuinely need FreshBooks Classic + Manychat legacy + an obscure Indian CRM with no public API, Zapier may be the only platform that has all three pre-built. Audit the connectors honestly before committing.
The 5-step decision tree
Run through this list in order. Stop at the first "no" and stay on Zapier.
- Do you do more than 2,000 automation operations a month? Below this, Zapier Free or Starter is cheaper than your time on Docker. Stay on Zapier.
- Can you run a single Docker container and remember to update it once a month? If no, stay on Zapier. The box will rot and you will lose more than ₹65K/year in dropped webhooks.
- Does n8n have first-party nodes for ≥80% of your integration count? Check at n8n.io/integrations. If yes, proceed. If no, calculate the cost of writing custom HTTP nodes for the gap and add it to the n8n side of the ledger.
- Are you certain you do not need SOC 2 / HIPAA in the next 12 months? If procurement is asking for it, stay on Zapier — DIY compliance costs more than the SaaS premium until you cross $500K ARR.
- Will you run [Hetzner](https://www.hetzner.com/) (EU, ₹580/mo) or an Indian VPS like [E2E Networks](https://www.e2enetworks.com/) (Mumbai, ₹800/mo)? Pick based on customer data residency. EU is fine for Indian B2C in 2026 (no DPDP transfer block yet). Mumbai is mandatory for fintech under RBI data localization.
If you cleared all five, set up n8n today. The whole migration from Zapier to n8n for a 10-Zap account took me 4 hours over a weekend. Export workflows as JSON from Zapier, recreate in n8n, point webhooks at the new endpoint, retire the Zapier Zaps in batches over 7 days.
For Indian MVPs specifically, if you are still on the 6-week MVP path, build the automation layer on n8n from day one. The ₹65K/year savings will fund your first marketing test. And if you need help wiring this into your stack — the founding engineer route handles automation, AI agents, and the Docker box together so you don't end up with three tools to maintain.