Everything here is free and needs no key and no sign-up. We collect each platform's public list of open roles at least once a day, check every role and publish the result in four ways: a search API, an MCP server for AI assistants, files you can download, and a widget you can put on your own site. Our own figures are open under CC BY 4.0. The roles belong to the platforms that post them, so the role data is for personal job search only.
What is open
| Address | What it holds | Terms |
|---|---|---|
/api/jobs/search | Search of the roles open today, JSON | Personal job search, not for redistribution |
/mcp | MCP server with the same search, for AI assistants | Personal job search, not for redistribution |
/jobs.json | Today's list, role by role | Personal job search, not for redistribution |
/data/jobs-public.json | The same list made for other tools, with no referral flag | Personal job search, not for redistribution |
/data/changes.json | The roles that opened and closed in the last 7 days | Personal job search, not for redistribution |
/pay-index.csv | AI Training Pay Index: listed hourly pay by platform and by field | CC BY 4.0 |
/country-index.csv | Country Job Index: open roles and listed pay by country | CC BY 4.0 |
/data/history-daily.csv | Daily totals per platform since the record began (also as JSON) | CC BY 4.0 |
/data/open/ | Monthly figures by platform, field, language and country | CC BY 4.0 |
/widget | The newest roles in a frame for your site | Free to embed, with its credit line |
/llms.txt | The main pages in plain text, for AI assistants | Free to read |
Every file is rebuilt with each daily read; /build.json gives the time of the newest collection of roles (read_at). Mercor and Terac are left out of /data/jobs-public.json, the API, the MCP server, the widget and the CC BY files, because their terms name benchmarking or aggregation; their roles stay on the site.
Jobs API
GET https://aiworkcentral.com/api/jobs/search answers JSON with the roles open today. One result is one role: a role a platform posts once per country comes back once, with each copy (its own link, countries and pay) inside copies. With country, the copy open in that country comes first.
| Parameter | What it takes |
|---|---|
q | Words that must all appear in the title or the tags, at most 100 characters. |
country | A two-letter country code (ISO 3166-1, such as US, BR or PT): only roles a person there can apply to. The platform's own list of countries it cannot hire in is applied first, then the countries the listing names; roles open worldwide count. |
lang | A work language (ISO 639-1, such as pt or de). en also gives the roles that name no language. |
field | One of coding, stem, writing, languages, medicine, law, finance, robotics, trades, general. |
min_pay | Lowest typical listed pay, in US dollars an hour, 0 to 1,000. |
since | YYYY-MM-DD: roles first seen on or after that day, at most 13 days back. |
since_hours | Roles first seen in the last N hours, 1 to 312, rounded down to the day. |
sort | new (newest first, the default) or pay (highest typical listed pay first). |
limit | Results to return, 1 to 50; 20 when left out. |
Examples:
curl 'https://aiworkcentral.com/api/jobs/search?country=PT&field=languages&limit=5'
curl 'https://aiworkcentral.com/api/jobs/search?q=python&min_pay=40&sort=pay'
curl 'https://aiworkcentral.com/api/jobs/search?since_hours=24&country=IN'
Each result has title, platform, platform_slug, field, first_seen and copies; each copy has url (our role page), pay (min, max, unit, currency, typical_usd_hour, as listed), countries, languages, date_posted and pay_index, which compares the listed hourly midpoint with the site-wide and field medians of the AI Training Pay Index of the same day. A wrong parameter answers 400 with param and a sentence that says what it takes.
MCP server
https://aiworkcentral.com/mcp is a remote MCP server over Streamable HTTP, with no sign-in. Its tools wrap the same search: search_roles, new_since, role_details, pay_index, market, cv_match. All of them only read. cv_match uses the CV text for that one answer: it is not stored, logged or kept.
In Claude Code:
claude mcp add --transport http aiworkcentral https://aiworkcentral.com/mcp
In a client that takes a JSON list of servers:
{"mcpServers": {"aiworkcentral": {"type": "http", "url": "https://aiworkcentral.com/mcp"}}}
Limits and caching
- The API and the MCP server share two limits: 3,000 calls an hour from one address, and 15,000 calls a day from all callers together. Over either, the answer is 429 with
Retry-After; the daily limit opens again at 00:00 UTC. - An API answer may be cached for 5 minutes, /jobs.json and the files under /data/ for an hour. The roles change once or twice a day, so polling more often than every few hours brings nothing new.
- From a script on another site, the browser can read the API, the MCP server and
/jobs.json,/data/changes.json,/data/history-daily.csv,/data/history-daily.json,/data/jobs-public.json. Other files download normally but are not open to scripts of other sites; read them from a server. - We keep nothing about callers: no address is written down, and the only number stored is the day's total of calls.
The role files
/jobs.json is the list this site shows, role by role, one entry per listing. /data/jobs-public.json is the same list made for other tools: no referral flag, a fields block that explains each field, and the terms written into the file. Fields of an entry:
| Field | What it holds |
|---|---|
id | Our id of the role: the platform slug and the platform's own id. |
platform | Platform slug; its name is in "platforms". |
title | The title the platform lists. |
url | The role page on aiworkcentral.com: the plain role link, with the apply link of the platform on it. |
path | The same page, as a path. |
referral | Only in /jobs.json: true when the Apply link of the role on our page is a referral link. |
family | Only on a role posted once per country: the id of its oldest posting. Entries with the same family are one role on the site, posted with a different page or pay per country; apply from the entry of your country. |
copies | Only when the postings of one role that say the same (pay, languages) and fit inside this entry's countries were folded into it: how many postings this entry stands for. |
bulk | Only on a role first seen in the last 14 days that came in a bulk load (a source read for the first time, or a mass copy) and was not posted that day or the day before: it is not counted as a new role. |
pay_min | Lowest listed pay, or null. |
pay_max | Highest listed pay, or null. |
pay_unit | hour, week, month, year or task, as listed. |
pay_once | Only when true: the pay is paid once (a bonus, a fixed project). |
pay_audio | Only when true: the pay is per hour of audio worked on, not per hour of work (pay_unit is task). |
currency | ISO 4217 code of the listed pay. |
pay_typical | One typical hourly figure in US dollars from the listed pay, or null. |
pay_basis | How pay_typical was read: range_middle (the middle of a range), max_only ("up to"), single. |
pay_for | What the pay is for when the listing says so, or null. |
pay_estimate | true when pay_typical is our estimate from a listing that gives no hourly rate. |
field | Our field slug (aiworkcentral.com/fields/<slug>). |
languages | Languages the role asks for, as ISO 639-1 codes (en included); empty when the listing names none. |
countries | Countries the listing names, in English ("Portugal"), or "worldwide"; empty when the listing names none. |
elig | Only on a role whose own text requires one country that its countries contradict: where it counts as open (that country, or "none": nowhere until the platform says otherwise); countries stays as listed. |
need | With elig only: what the listing requires, in English (e.g. "US medical licence and US residence required"). |
location | The location text of the listing. |
employment_type | Contract type as listed, or null. |
hours | Only when the listing says: the hours line, in English. |
date_posted | The date the platform gives as posted, or null. |
valid_through | The end date the platform publishes (YYYY-MM-DD), only when it is today or later; for micro1 it is set automatically 60 days after posting, not an announced closing date. A past one is left out. |
first_seen | The day we first saw the role, while under 14 days old; null after. |
summary | Our own short summary of the listing. |
tags | Skill words found in the title or summary, in English. |
locales | Only when the title names language variants: their tags (see "locales" at the top). |
Open data under CC BY 4.0
Our own figures, worked out from the roles: counts, medians, quartiles and day counts. No role, title or link is in these files.
| File | Columns |
|---|---|
/pay-index.csv | edition, |
/pay-index-transparency.csv | edition, |
/country-index.csv | edition, |
/data/history-daily.csv | date, |
/data/open/aggregates-YYYY-MM.csv | month, |
The method of each file is on its page: Pay Index, daily record, monthly figures, Country Job Index. Each monthly edition of the Pay Index and the Country Job Index is frozen once published and never edited.
What you may reuse, and how to credit it
The figures of the files above are published under CC BY 4.0: copy them, change them, use them commercially, with credit. Cite a Pay Index edition like this:
AI Work Central, AI Training Pay Index, October 2026 edition (figures as of 2026-10-30), https://aiworkcentral.com/pay-index. Derived figures licensed CC BY 4.0.
For anything else from this site, the credit line is Source: AI Work Central (aiworkcentral.com), with a link to the page the data came from.
Not licensed: the roles. Titles, summaries, descriptions, pay text and the links of a listing belong to the platform or employer that posts it, and some platforms' terms forbid copying or aggregating them. So /jobs.json, /data/jobs-public.json, /data/changes.json and the answers of the API and the MCP server are for personal job search: finding roles for yourself, or showing a person the roles that fit them. They are not licensed for redistribution: not for a copy of the list, and not for a job board built on it. Platform names are their owners' names and are not licensed either. The widget below is the way to show our roles on your site.
Widget for your site
A small frame with the newest open roles, for a blog, a community page or a careers page. It loads no script, sets no cookie and counts nothing on your page. Each role links to its page on AI Work Central, where the reader sees the details and applies on the platform. Roles come newest first, the same order for every platform; the widget has no paid or Featured places.
<iframe src="https://aiworkcentral.com/widget?country=US&field=languages&n=5" title="Latest AI training roles, from AI Work Central" width="100%" height="520" style="border:0" loading="lazy"></iframe>
| Parameter | What it takes | When left out |
|---|---|---|
country | A two-letter country code (US, BR, PT): only roles a person there can apply to. | Every role |
field | One of coding, stem, writing, languages, medicine, law, finance, robotics, trades, general. | Every field |
n | How many roles to show, 1 to 10. | 5 |
theme | light or dark. | The reader's device setting |
Examples: roles open in the United States, language roles, dark, coding roles open in India. Keep the "Powered by AI Work Central" line visible; it is the credit the widget asks for.
For AI assistants
/llms.txt lists the main pages of the site in plain text, with the API and the MCP server, for assistants and crawlers that read it.
Something missing or wrong? Tell us on the contact page.