Coding agents and MCP
A backend your coding agent can build on, test and ship by itself.
Connect Claude Code, Cursor or any MCP client with one command. Your agent writes the app contract and React screens, applies them to a real Postgres backend, runs verify, reads the screenshots and deploys, without you wiring anything.
>Add Google sign-in and a rule: no double-booking.
appbricx.patch_contract auth.google, 1 rule
appbricx.verify
role matrix: 3 roles × 4 tables
every page opened as every role
appbricx.screenshot 8 pages
appbricx.deploy
live at clinicly.apps.appbricx.com
- 24 MCP tools
- From create_project to deploy, sql and secrets
- Browser sign-in
- OAuth in a browser window, no token to paste
- CLI and HTTP API
- The same operations for scripts and CI
- No agent?
- The built-in builder works from a description
How your agent works with Appbricx
Add the MCP server. In Claude Code it is one claude mcp add command; Cursor and other clients take a block of settings. A browser window asks you to authorize it.
Describe the app in plain words. The agent reads get_guide, writes the contract (tables, roles, pages, rules, workflows) and your own React components, then calls check_contract and apply_contract.
The agent runs verify, looks at the screenshot of every page as every role, fixes what failed and calls deploy. You get a live link.
# connect Claude Code to Appbricx
claude mcp add --transport http appbricx https://appbricx.com/api/mcp
# a browser window asks you to authorize it
# then, inside Claude Code
> Build a booking app for my clinic. Patients book,
doctors keep notes, and no one gets double-booked.
● appbricx.apply_contract 4 tables, 3 roles, 1 rule
● appbricx.verify static, data and browser checks
✓ every page opened as every role
● appbricx.deploy
✓ live at clinicly.apps.appbricx.com
Everything an agent needs, as tools
The contract loop
get_contract, check_contract, apply_contract and patch_contract change tables, roles, pages and workflows in small, checked steps.
Checks it can read
verify returns a report that names the role, table or page and a fix; screenshot returns every page as each role.
Your own screens
check_component type-checks a React component before it ships. Screens use the @app/kit SDK for data, sign-in, uploads, payments and AI.
Operate production
sql, edit_rows, describe_schema, snapshots, deploys, run, tail and secrets work on dev or prod, with a snapshot before every prod change.
Scoped access
Reads need read, dev writes need apply, prod writes and rollbacks need deploy. You choose the scopes when you authorize the agent.
The same API everywhere
The CLI and the developer HTTP API with personal access tokens run the same operations for scripts and CI.
Learnly, built live by Claude Code
This is the real Claude Code session that made Learnly, condensed: contract, checks, a failed verify and its fix, then deploy. The app is live with demo logins for a student and an instructor.
Coding agents and MCP: common questions
Which coding agents work with Appbricx?
Claude Code, Cursor and any other client that supports the Model Context Protocol. Scripts and CI can use the CLI or the HTTP API with a personal access token instead.
How does the agent sign in?
When you add the MCP server, a browser window opens and asks you to authorize the agent with the scopes you choose. There is no token to copy.
Do I pay for my agent's model usage twice?
No. When your own agent builds the app, its model usage stays on your provider bill and Appbricx uses no AI credits. Your plan pays for the apps you run.
What if I don't use a coding agent?
Use the built-in builder: describe the app in plain words and it builds the same kind of app, on plan credits or your own model key.
Can the agent write its own React screens?
Yes. The agent writes React components, checks them with check_component, and they ship with the app. A brand colour and a custom sign-in screen are part of the contract.