Connecting tools without writing a backend

Ten years ago, "I want X to talk to Y" meant standing up a backend service, hosting it somewhere, monitoring it, paging yourself when it broke at 2am, and explaining to your co-founder why the Stripe-to-CRM glue had its own on-call rotation. In 2026 the answer is almost always one of three patterns and none of them require writing a server.

Most "we need a backend for this" instincts are muscle memory from a different decade. Check the easier paths first.

The three patterns, in order of preference

Native integrations. The two tools already know about each other. Stripe to HubSpot. Slack to Linear. Gmail to Notion. Calendly to Salesforce. Open the integrations tab in either tool. If it exists, click the button, authorize, map the fields, done. Ten minutes. No URLs, no keys you have to rotate, no code that becomes someone's problem when they leave. The product team at both vendors is paid to keep this working. Let them.

Webhooks. Tool A pushes to a URL when something happens. The URL is owned by Zapier, Make, n8n, or a 20-line serverless function. Nothing is "running" between events. There's no server idling, no container to restart, no health check to write. A function fires when called and goes back to sleep. For most "X happens, do Y" flows, this is the shape.

Polling-and-glue. When neither of the above exist, a scheduled tool checks every five minutes, calls API A, calls API B, glues the result. This is what Zapier scheduled triggers and Make scenarios are doing under the hood. Slower than webhooks, but it works against any tool with an API, even ones whose product teams have never heard the word "outbound."

A decision tree

Walk these in order, stop at the first yes:

• Does a native integration exist between the two tools? Use it.
• Does Tool A support outbound webhooks? Point them at a Zapier or Make endpoint, no code.
• Does Tool A have an API but no webhooks? Set up a polling Zap or a scheduled n8n flow.
• All else fails? A 30-line serverless function on Vercel cron, Railway cron, or a Cloudflare Worker on a schedule.

Notice "real backend service" doesn't appear on the list.

The "do I actually need a backend" test

You probably don't. The honest exceptions:

• Latency matters and the user is waiting (under one second response time, in-app)
• Volume is real (10k plus events per day, where Zapier task pricing breaks your budget)
• You're stitching real-time flows that touch your own database and your own auth
• Compliance requires the data not leave your control

If none of those apply, glue it.

A concrete walkthrough

"Every new Stripe customer should land in HubSpot with the right tags." Four steps in Zapier or Make:

1. Trigger: New Customer in Stripe.
2. Filter: Only if customer has paid (skip trial signups, or don't, your call).
3. Action: Find or create contact in HubSpot, mapping email, name, plan, signup date.
4. Action: Add tag based on plan tier ("starter," "pro," "enterprise").

Test with one real customer. Turn it on. Total elapsed: about ten minutes. Total code: zero. You will not remember this exists in six months, which is the point.

The hidden trap

Glue tools break silently. A vendor changes their API, your Zap stops firing, you find out three days later when a customer asks why they didn't get added to your CRM. The flow that "just worked" for a year suddenly hasn't worked all week.

Two defenses. First, set up a "this is healthy" check. Better Stack ping, or a scheduled rule that says "if no events have flowed through this Zap in 24 hours, alert me." Second, commit to checking your Zap or scenario history once a week. Five minutes. You're looking for the run that errored, not for green checkmarks.

The trap isn't that glue tools fail. It's that they fail without telling you, and the failure mode is "nothing happens," which is indistinguishable from a quiet week.

Cost

Zapier runs $20 to $50 a month for a typical small-team setup. Make is meaningfully cheaper per operation if you're hitting volume. n8n self-hosted is free if you've already got infrastructure to run it on, and the overhead is real but bounded.

The math, almost always: anything you'd spend less than a day building and maintaining yourself, just glue. Your time costs more than the subscription. The day you write your own service is the day you've taken on a maintenance burden that compounds. Glue tools have one job and a team paid to keep them working.

Up next

The middle pattern (webhooks) and the third (polling) look interchangeable until you ship something real. The next guide, "Webhooks vs polling: when each one wins," covers when to reach for which, and why the answer is rarely "whichever is easier to set up."