Why Baseboard
Why not just build it with AI?
You can. A coding agent will give you a users table joined to Stripe in an afternoon. This page is about what comes after that afternoon, so you can decide with the whole bill in front of you.
The bill
The first version is the cheap part
What it takes to run an admin panel over the services you already use, built yourself and in Baseboard.
| The job | Building it yourself | In Baseboard |
|---|---|---|
One record per customer Sign-in users joined to billing customers | A join on email, then fixes for case and stray spaces, guest checkouts, deleted customers and every page of results past the first. | Joined on every read, across every sign-in and billing connection you add. |
Each provider’s API Pagination, rate limits, OAuth | Written once per provider and kept working as their APIs change. OAuth tokens need refreshing before they expire. | 106 providers, kept up to date by us. OAuth tokens are refreshed for you. |
Keys to production Stripe, your database, your host | Live keys in your app’s environment, and in the agent’s context while it writes and tests the code. | Each key sealed on its own with AES-256-GCM. One broker opens them, and agents never see one. |
Who may do what Refunds, bans, rollbacks | Roles, a level per action and a confirm step. Usually added after the first mistake. | Everyone starts read-only. An owner allows each change, person by person. |
A record of what happened Audit log | An audit table, and remembering to write to it from every code path that changes something. | Every read and every change logged. Agents are logged under the token they used. |
Agents and scripts MCP, tokens, a CLI | An MCP server, token scopes and limits per project, on top of everything above. | 67 tools over MCP, with read, change and setup scopes and a CLI. |
What goes wrong
Where homemade admin panels hurt
None of these show up on day one. They show up in a support ticket, or in an incident review.
It’s written with the keys in hand
To test against real data the agent needs live keys. They end up in .env files, shell history and its context.
Internal code gets the least review
It sits behind a login, so it’s checked less than the app your customers use. It’s also the code that can refund and delete.
Everyone is an admin
The first version has one role. That lasts until you hire support and want them to look things up without issuing refunds.
Nobody owns it after the first week
It works, so nobody opens it again until a provider changes its API or your customer list passes the first page.
Being fair
When you should build it yourself
Baseboard isn’t the right answer for every team. Build your own if one of these is you.
Your admin work is your own logic
Approving listings, moderating posts, running payouts: work that lives in your code rather than in Stripe or Clerk.
You need one table and nothing else
A list of rows from your own database is a good job for a coding agent.
Everything must run on your own servers
Baseboard is a hosted service. If your rules don’t allow that, it won’t fit.
Or do both
Build the screens only you can write, like approving listings, in your own app. If your users, billing, tables or files live there too, serve them from a Custom API endpoint and Baseboard shows them next to your other providers. You build the part that’s yours and skip maintaining the parts that aren’t.
Questions
What people ask next
Can’t I give Claude or ChatGPT my keys and ask it questions directly?
What if I need something Baseboard doesn’t do?
What if we stop using Baseboard?
npx baseboard-cli pages export, then revoke the keys at each provider.Is it cheaper than building it?
See what it finds before you build anything.
Free reads every page and changes nothing. Connect a sign-in provider and a billing provider and look up your first customer.