Extra tools you can let the agent use
Optional outside tools the agent can use while building your app, and how to switch them on.
MCP tools is a list of optional outside tools the agent can use while it builds your app. Each one is started only when it is used, and only for the one app you turned it on for. Turning one on is your decision alone: the agent cannot switch one on itself.
You find the list in the app's Settings, under MCP tools.
What is available
One tool comes with AppHandle today.
- GitHub lets the agent read and work with repositories, issues and pull requests. It needs a GitHub token, which you store in the app's Settings → Secrets.
A tool that is on but still missing something says so: the row shows Needs: followed by the name, with a link that takes you to the app's stored credentials. The tool stays unusable until you add the value, and then it shows Ready to use.
Turning a tool on or off
Each row has one button that reads Enable or Disable. While a tool is off, its row says Off — not available to the agent. When it is on, the row shows Active.
Turning one off takes effect immediately: the agent stops being offered that tool.
Add your own MCP server
If you have an outside tool that follows this standard, select Add your own MCP server. You give it a name, an address, and an optional description. The address has to be a public one that starts with https, and only this app's agent can reach it.
If the server needs a key, add a header, choose Secret, and pick one of the app's stored credentials. The key is kept as a credential rather than typed into the app. You can also choose Literal for a value that is not secret, or Integration to use an account you have connected.
Once added, your server appears in the list with a Custom badge, and you can Edit or Remove it.
What it costs
Calls to a server you add yourself are metered — from your monthly included usage first, then your balance — at a fraction of a cent each; current prices are on the pricing page. The tool that comes with AppHandle is not metered this way.
When a tool is not working
Under the list there is a Recent diagnostics block. It records what happened to each tool — connected, failed to start, a call that failed after reconnecting, or output above the size limit — with the time, so you can see why one is not working.
Where to go next
Teaching the agent new tricks covers the skills you can add, and Connect an outside API covers calling a service directly from your app.