Build a REST API Without Writing Code

"Build a REST API" sounds like a job for a backend engineer with years of framework experience. It used to be. The unglamorous truth is that most small APIs are extremely mechanical: accept some data, store it, hand it back out, tell the caller whether each step worked. Mechanical work is exactly what AI coding agents are good at, which makes a first API one of the best projects you can attempt without writing code yourself. This is how to do it so the result actually works, not just looks like it does.

What a small API actually consists of

Strip the jargon and a typical beginner API does four things. It saves something (a booking, a task, a recipe), lists what's been saved, hands back one item by its ID, and changes or deletes items on request. Data moves as JSON — plain labeled text any language can read — and every request gets back a status code: 200 means fine, 404 means not found, 400 means you sent something malformed. That's genuinely most of it. Authentication, pagination, and caching are real concerns eventually, but none of them are required to ship something useful on day one.

Start with the data, not the code

The highest-leverage thing you'll do has nothing to do with endpoints. Before touching the agent, write down what one record looks like. A booking might be: an ID, a customer name, a date, a time slot, and a note field. Decide which fields are required and which are optional. This half-page of thinking becomes the core of your prompt, and it prevents the single most common beginner failure: an API that stores things loosely and can't answer questions about them later.

If your data currently lives in a spreadsheet, that sheet is your schema draft — columns become fields, rows become records. The same path works whether you're turning it into an app or an API behind one.

Describe the endpoints in plain English

Now brief the agent. Name the resource ("bookings"), then list the operations: create one, list all, get one by ID, update one, delete one. Say who will call this API — your own website, a mobile app, a script — because that changes defaults like error formats. State the required fields from your schema and ask the agent to reject requests missing them. If anything should be private, say so explicitly; "add a simple API key check" is a complete sentence, and agents handle it fine.

One habit separates good outcomes from mush here: describe behavior, not implementation. Don't ask for Express or FastAPI or Postgres unless you have a real reason. Ask for "a small JSON API with these five operations," and let the agent pick boring, well-supported tools.

Test it before you trust it

An agent will happily report success while the server fails to start. Verify with curl, the command-line tool for poking APIs:

curl http://localhost:8000/bookings
curl -X POST http://localhost:8000/bookings -H "Content-Type: application/json" -d '{"customer":"Kim","date":"2026-09-01"}'

Check three things specifically. Does creating an item actually return the new record with an ID? Does requesting a nonexistent ID return 404 rather than crashing the server? And does sending garbage — an empty body, a date as free text — produce a clean error instead of silent nonsense? Ask the agent to fix whatever misbehaves, re-run the same curls, and only move on when all three pass. Five minutes of this teaches you more about how the pieces fit than an hour of reading.

Put it somewhere real

localhost is your machine talking to itself; nobody else can reach it until the API runs somewhere public. Deploying an API is the same motion as deploying any AI-built project — get the app built with AI onto a real host — with two extra checks. First, environment variables: database locations and API keys must be configured on the server, not just on your laptop, or the deployed copy dies on first contact. Second, confirm the deployed URL answers the same curls that passed locally. If they pass against the live URL, you have a working API, full stop.

The meshcode angle

Building and testing an API is a loop — prompt, run, curl, fix — and loops go faster when the terminal running the server sits next to the agent editing the code. meshcode is a native desktop app for macOS and Windows where panes hold separate agent sessions over the same repo: one pane keeps the dev server and curl tests running while another iterates on the code. Drive the Claude Code or Codex CLI subscription you already pay for, or start with meshcode's own metered models billed pay-as-you-go.

👉 Download meshcode — Mac, Windows