How to automate a website that has no API
Use the interface a site already ships when there is no endpoint: vendor portals, admin consoles, legacy systems and internal tools an agent can operate directly.
When a site has no API, the interface is the integration surface. A browser agent reads the page, performs the same actions a person would, and verifies the result, which makes vendor portals, admin consoles and legacy internal tools automatable without waiting for an endpoint that may never exist.
The integration that never arrives
Most business software has partial API coverage at best. The workflow you need is often the one behind a table, an admin screen, a supplier portal or a legacy internal tool, and the endpoint is either missing, undocumented, rate limited, or gated behind a tier you do not have.
The browser is the common layer across all of it. Anything a person can do in a tab is automatable, which makes the browser the integration surface of last resort and, often, the only one available.
- Supplier and partner portals with no public API
- Admin consoles where the API omits the exact action you need
- Legacy internal systems nobody will re-platform
- SaaS tiers that put API access behind a higher plan
Doing it responsibly
Browser access is not permission to ignore a site’s rules. Use an agent only where you are authorised, and respect terms of service, robots instructions where applicable, rate limits, privacy obligations and applicable law. This is a tool for legitimate work, not a way around access controls or anti-bot protection.
In practice that means: your own accounts, your own vendor portals, systems your organisation is entitled to use, and a human reviewing anything consequential.
Making it reliable
Give the agent the narrowest scope that completes the job, start from the page rather than asking it to find the site, and require approval before writes. Verify the result the way a person would: read the page back after the action and check the change actually happened.
- Start on the tab that already holds the record
- Keep scope at This tab unless the task genuinely spans several
- Approve before submit, publish, pay or delete
- Ask for the evidence, not just the conclusion
Frequently asked questions
Is this the same as web scraping?
Related but broader. Scraping reads. A browser agent can also act: fill a form, update a record, move information between systems. Both carry the same responsibility to respect terms and law.
Will it break when the vendor redesigns the portal?
Far less often than a selector script, because the agent targets meaning rather than DOM structure. A redesign that would break a script is usually just a re-read for an agent.
Can it work behind a login?
Yes. It runs in your real browser on a tab you attach, so it uses the session you are already signed into. It never types passwords.
