For the impatient:
- Prepare an environment by either:
- Run the application
This repository contains an example bank ("Rebank") written using Reboot, with its API defined using Pydantic models.
The Reboot '.proto' definitions
can be found in the api/ directory, grouped into
subdirectories by proto package, while backend specific code can be
found in backend/ and front end specific code in frontend/web/ and MCP Apps in frontend/mcp/ and a React Native (Expo) mobile front end in frontend/mobile/.
For more information on all of the Reboot examples, please see the docs.
This method requires running VSCode on your machine: if that isn't your bag, see the other environment option below.
This repository includes a Dev Container config (more about Dev Containers) that declares all of the dependencies that you need to build and run the example. Dev Containers can be started locally with VSCode, but we recommend using GitHub's Codespaces to quickly launch the Dev Container:
- Right-click to create a Codespace in a new tab or window:
- Go to https://github.com/codespaces and click the three dots next to the codespace you just created and then click
Open in Visual Studio Code.- You can set your default editor to VSCode for codespaces to avoid this step in the future. See these instructions for more information.
Now you're ready to run the application!
Running directly on a host requires:
- A platform of either:
- Linux:
x86_64orarm64, withglibc>=2.35(Ubuntu Jammy and other equivalent-generation distributions) - Mac:
arm64(Apple M1 and newer) withMacOS>=14.0andXcode>=15.2 - ... or talk to us if your desired platform isn't currently supported!
- Linux:
- uv - A tool to manage
python,pip, andvenv.- If you are already familiar with Python virtual environments, feel free to use your tool of choice with
pyproject.toml. Python >= 3.10 and < 3.13 is required.
- If you are already familiar with Python virtual environments, feel free to use your tool of choice with
- Node.js
- Including
npm.
- Including
- Docker
- Note: the example does not run "inside of" Docker, but Docker is used to host a native support service for local development.
If you are unable to meet any of these requirements, we suggest using the VSCode and Dev Container environment discussed above.
Now you're ready to run the application!
Our backend is implemented in Python and we must install its dependencies before
running it. The most notable of those dependencies is the reboot PyPI
distribution, which contains both the Reboot CLI (rbt) and the reboot
Python package.
Using uv, we can create and activate a virtualenv containing this project's dependencies (as well as fetch an appropriate Python version) using:
uv sync
source .venv/bin/activateThen, to run the application, you can use the Reboot CLI rbt (present in the active virtualenv):
rbt dev runRunning rbt dev run will watch for file modifications and restart the
application if necessary. See the .rbtrc file for flags and
arguments that get expanded when running rbt dev run. By default it
serves the app under a name, https://bank.localhost, through
portless: install it once
with npm install -g portless and start its proxy with
portless proxy start. To use plain ports instead, run
rbt dev run --config=ports.
Similar to the backend, the front end has dependencies that need to be installed before running it. Open a separate terminal/shell and do:
cd frontend/
npm install
npm run devThat starts a single Vite dev server that serves both the web app and
the MCP Apps, as https://frontend-bank.localhost, which the backend
serves them from (the name matches the backend's bank.localhost,
which is how the two trust each other): npm run dev is
portless frontend-bank vite, portless being a dev dependency,
mirroring the .rbtrc. Visit
https://bank.localhost for the web app.
Without portless (rbt dev run --config=ports), run
npm run dev:ports instead and visit
http://localhost:9991.
The web app requires signing in. In development, click "Sign in" and
pick (or make up) a Development identity. Signing in auto-constructs
your User, which signs you up as a customer of the bank (see
backend/src/user_servicer.py), so the web app shows only your own
accounts. The web app and the MCP frontend share sign-on: signing in on
one signs you in on the other.
The mobile app signs in too, against the same OAuth server, and with
the same useSignIn() / useSignOut() / useUser() hooks. It cannot
use the browser-redirect flow the web app uses — React Native has no
page to redirect and no cookie jar to hold the session — so it builds
an auth object with expoAuth({...}) from
@reboot-dev/reboot-react/native and passes it to its
RebootClientProvider as nativeAuth, and Reboot runs the standard
authorization-code flow with PKCE that native apps use instead,
keeping the resulting tokens in the device keychain. Because
backend/src/main.py lists the app's redirect URI in
OAuth(skip_consent_for_redirect_uris=...), Reboot signs the user
straight in, with no consent screen — the same treatment the web app
gets. A client that registers some other redirect URI still gets
one.
The application is also an MCP server: an MCP client can call
its tools and render its UIs (the frontend/mcp/ directories) right
in the chat. The easiest way to get connected is the setup wizard
that the backend serves at
https://bank.localhost (or
http://localhost:9991 without portless): it
walks you through connecting the chat client of your choice (Claude,
ChatGPT, ...) — including the public tunnel those clients need to
reach your machine — and suggests example prompts to try. The MCP
endpoint itself is
https://bank.localhost/mcp.
The application comes with backend tests.
Before you run the tests, you'll
need to ensure you've run rbt generate. If you've already run rbt dev run
without modifying .rbtrc, rbt generate will have been run for you as
part of that command.
Otherwise, you can do it manually.
rbt generaterbt generate will automatically make required Reboot '.proto'
dependencies like rbt/v1alpha1/options.proto available on the
import path without you having to check them into your own repository.
Now you can run the tests using pytest:
pytest backend/