Trawl
Describe the data. Get JSON that matches.
Give Trawl a sentence, a JSON Schema and, if you know it, the page to start from. A browser agent reads the web for you and returns JSON that fits, on a signed webhook.
Describe it, get JSON that fits.
You send a sentence and the JSON Schema you want back. The result is checked against your schema before it reaches you, so your code gets the shape it expects.
No scraper to maintain.
A browser agent reads the page the way a person would. When the site is redesigned, there are no selectors to fix, because there never were any.
Results on a signed webhook.
Trawl posts each finished job to your endpoint, signed so you know it came from us. Send test and replay are one click away.
Nothing is lost if your endpoint is down.
Every result stays on its job. If a delivery fails, fetch the result whenever you like, and reading a job is free.
How it works
Your first extraction in three steps.
- Turn on Trawl for your workspace and create an API key.
- Send a sentence and the JSON Schema you want back.
- Get the result on your webhook, or read it from the job.
import { Trawl } from "@productcraft/trawl";
const trawl = new Trawl();
// A sentence and the JSON Schema you want back. The page is optional:
// leave suggested_urls out and the agent searches for one.
const job = await trawl.jobs.createAndPoll({
description: "The five most recent releases: package name and version.",
json_schema: {
type: "object",
properties: {
releases: {
type: "array",
maxItems: 5,
items: {
type: "object",
properties: { package: { type: "string" }, version: { type: "string" } },
required: ["package", "version"],
},
},
},
required: ["releases"],
},
suggested_urls: ["https://github.com/clauderanelagh/productcraft-node/releases"],
});
console.log(job.status, job.result); // the result is checked against your schemaThe job above, as it ran: what went in, and the JSON that came back.
Describe it The schema and the sentence you send.
The five most recent releases: package name and version.
{ "type": "object", "properties": { "releases": { "type": "array", "maxItems": 5, "items": { "type": "object", "properties": { "package": {"type": "string"}, "version": {"type": "string"} }, "required": ["package", "version"] } } }, "required": ["releases"] }Get JSON The data the agent returned, checked against your schema.
{ "releases": [ { "package": "productcraft", "version": "0.3.1" }, {"package": "@productcraft/social", "version": "1.0.0"}, {"package": "productcraft", "version": "0.3.0"}, {"package": "@productcraft/waitlist", "version": "0.2.0"}, {"package": "@productcraft/trawl", "version": "0.2.0"} ] }
More in the box
Everything else Trawl does.
- Starts from a search No page to point at? Leave it out and the agent searches for one.
- Retries never queue twice Send with a key, and a retried request returns the first job instead of starting a second.
- Limits per job Cap how many steps and how long a job may take.
- Job history List jobs by status, and cancel one that is still running.
- Rotating signing secrets Each endpoint has its own signing secret, and rotating it is one call.
- Hosted in the EU Trawl runs in Frankfurt with the rest of ProductCraft.
For your AI agents
MCP tools to start a job, read its result and manage webhooks, behind the policies you set on each key, so an agent can gather data from the web and get it back as JSON. Read about MCP
Your first extraction, end to end.
Send a sentence and a schema, then read the JSON from the job or your webhook.