On mobile?
Email yourself the setup link so you can finish when you're at your computer.
✓ Email sent! Check your inbox.
× Failed to send. Please try again.
⏱ Too many attempts. Please wait a few minutes and try again.
Connect Claude, Cursor, ChatGPT or Copilot to AppAmbit and your agent stops describing what you should do next. It creates the database, deploys the function, flips the flag and reads the live crashes itself, through 34 tools scoped to your team. You stay in your editor, and everything it touches is waiting for you in your dashboard.
Works with App Database, Cloud Code and push notifications. No integration code to write.
create-database-tool
write
Database created and linked to your app
execute-database-statement-tool
guarded
Table reviews added: product, rating, comment, author
create-cloud-function-tool
write
Function post_review created
deploy-cloud-function-version-tool
write
Version 1 built and ready to go live
activate-cloud-function-version-tool
write
Live at POST /api/fn/post_review
Your agent calls the tools. AppAmbit does the work.
Tools your agent can call, across your whole workspace.
Read-only. Most of the surface looks, and changes nothing.
Endpoint. One URL and one token is the entire setup.
MCP client. Claude, Cursor, ChatGPT, Copilot, Codex.
These are ordinary sentences, not commands you have to learn. Your agent picks the tool, runs it against your workspace, and tells you what came back. Every one of these results is something you can open in your dashboard a second later.
Why did version 3.2 start crashing yesterday?
→ list-crashes-tool(app_key, start_date: "yesterday")
7 groups · top one hit 412 users in CartView
Turn the new checkout off for Android below 2.4.
→ update-app-config-tool(key: "checkout_v2", enabled: false)
Updated and live. No rebuild, no store review
Give this app a database with a products table.
→ create-database-tool then execute-database-statement-tool
app_db active · products created and queryable
Deploy a function that pushes an order confirmation after checkout.
→ create → deploy → activate-cloud-function-version-tool
Live at POST /api/fn/order_confirmed · it sends the push
How did the app do last week? Sessions and crash-free rate.
→ query-sessions-tool(app_key, start_date, end_date)
Totals, active right now, average duration, day by day
Add an FAQ section my app can read without a release.
→ create-content-type-tool then create-content-entry-tool
Published, and your SDK picks it up right away
The server groups its tools the way your backend is actually shaped. Your agent discovers all of it on the first call, then works with whichever part your question is about.
Read crashes and sessions, list and change remote config, and send a push, on any app in the team.
Create a database, link it to an app, inspect the schema, run SELECTs, and write with a guarded statement tool.
The full function lifecycle: read the handler contract, create, deploy a version, activate it, run it, read the logs.
Define content types, publish them so the SDK can discover them at once, and create or query entries.
Generate the SDK setup code for a platform, then check whether the integration is really reporting.
One call answers why a feature looks disabled: the plan, the feature flags, the limits and the current usage.
This is the exact list your agent sees when it connects, with the safety marking each tool carries. Read tools only look. Write tools change something. Guarded tools can remove data, so they ask before they do.
There is no SDK to install for this and no service to run. Point your assistant at the endpoint below with a team-scoped token and it discovers every tool on its own.
In your dashboard, open Setup and create a team-scoped token. It is shown once, so copy it there and then.
Drop the config into your assistant. One URL and one Authorization header is the whole integration.
Your agent calls get-started-tool, sees your apps and your plan, and goes straight to work.
{
"mcpServers": {
"appambit": {
"type": "http",
"url": "https://appambit.com/mcp/appambit",
"headers": {
"Authorization": "Bearer <your-token>"
}
}
}
}
Copilot uses servers instead of mcpServers. On Cursor the type line is optional.
[mcp_servers.appambit] type = "http" url = "https://appambit.com/mcp/appambit" [mcp_servers.appambit.http_headers] Authorization = "Bearer <your-token>"
https://appambit.com/mcp/appambit
Add it as a custom connector and approve the sign in. Nothing else to configure.
No account yet? Your token is in the dashboard the moment you are in. Full setup is in the docs.
Handing an agent write access to your backend should not feel like a leap. Every tool declares what it can do, so your client knows which calls are harmless and which deserve a prompt first.
A token carries a single team. Every tool resolves apps inside that team, and app keys from another team are simply not reachable on that connection.
21 of the 34 tools cannot change anything. They are marked read-only, so your client can auto-approve them and still stop to ask before the rest.
Four tools that can destroy data are flagged destructive, and an irreversible DROP or DELETE is refused unless the call passes an explicit confirm.
Revoke the token from your dashboard and every assistant holding it loses access on the next request. No redeploy, no cleanup.
Your agent stops handing you a checklist. It calls the tool, and the database, function or flag actually changes.
Database, cloud functions, content, remote config, push, crashes and sessions, all behind a single URL.
No dashboard hopping and no context switch. Ask in the editor you already have open and keep going.
AI App BuilderThe create tools take arrays, so ten tables or ten functions are one call, and per item results show what landed.
Flip a flag, send a push, roll a function back to yesterday. Things you would normally click, asked for in a sentence.
Remote configEverything the agent creates shows up in AppAmbit, so you can review a function, browse a table or read a crash yourself.
Push notificationsStart free, generate a token, and give your agent something real to build in the next five minutes. No credit card required.
Your agent runs the backend.