Skip to content
Magic Coder
Powered by BridgeApp

Overview

Magic Coder is a coding agent. You start it from your terminal, point it at a repository, and ask it to do work — fix a failing test, refactor a module, write a feature, investigate a bug. It reads your code, edits files, runs commands, and keeps going until the task is done.

Magic Coder is part of the BridgeApp ecosystem. BridgeApp is the platform — your account, your data, the agentic engine running the model. Magic Coder is the surface that runs on your machine. The two are tightly coupled: you sign in once with a BridgeApp account, then Magic Coder uses that identity to run sessions, persist threads, switch models, and access your BridgeApp tasks and docs.

Install

Section titled “Install”

Homebrew (macOS and Linux):

Terminal window
brew tap MathAndMagic/magic-coder
brew install magic-coder

Or grab a binary directly from the releases page. See Install for the full matrix.

Start your first session

Section titled “Start your first session”

From the root of any repo:

Terminal window
coder

The first time you run this, Magic Coder:

  1. Opens a browser window to sign in to BridgeApp (one-time per machine).
  2. Asks whether to trust the workspace. Trust enables project-local rules and skills, and lets you opt into hands-off --automagic runs later.
  3. Drops you into the TUI with a prompt. Type a request and press Enter.

A typical first ask:

fix the failing tests in src/lib/parser.rs

Magic Coder reads the relevant files, runs the test command, applies a diff, re-runs, and reports back. Approve each shell command and edit as it goes — or pick a hands-off mode (see below).

For a guided walkthrough, see Your first session.

What Magic Coder can do

Section titled “What Magic Coder can do”
  • Read your repository. Lists directories, reads files, greps for patterns, follows references.
  • Edit files with diffs. Changes are applied as patches, line by line, with approval.
  • Run shell commands. Build, test, run a script, install a dependency, start a dev server in the background.
  • Plan first. In plan mode, Magic Coder investigates and proposes a concrete implementation plan before touching anything.
  • Run hands-off. Automagic auto-approves every prompt and exits when the task is done — for known shapes like fixing tests or regenerating artifacts.
  • Resume anywhere. Sessions are stored as threads. Pick one up on a different machine with coder --continue <uuid>.

What you control

Section titled “What you control”

Magic Coder is opinionated about safety:

  • It can only read and edit files inside the workspace roots you point it at — usually the repo you launched it from.
  • The first time you start it in a directory, you decide whether to trust the workspace. Until trusted, project-local rules don’t apply and most operations require approval.
  • Shell commands prompt for approval before they run. You can approve once, remember the decision for this workspace, or run with --automagic for a non-interactive flow.

You can change the model, reasoning effort, and long-context mode at any time mid-session with a slash command.

Where to go next

Section titled “Where to go next”