@mushi-mushi/cli

@mushi-mushi/cli

Your AI wrote it. Mushi tells you why it broke.

The mushi command for Mushi Mushi. Install the SDK in your app, connect your editor, then read bugs, dispatch fixes and merge them without leaving the terminal.

flowchart LR
    A["mushi init<br/>SDK in your app"] --> B["mushi setup<br/>editor reads diagnoses"] --> C["mushi reports<br/>see what broke"] --> D["mushi fix<br/>agent opens a draft PR"] --> E["mushi fixes merge"]

Quick start

npx mushi-mushi                      # the SDK wizard (same as: npx @mushi-mushi/cli init)
npx mushi-mushi setup --ide cursor   # connect your editor: cursor, claude, continue or zed
npm install -g @mushi-mushi/cli      # optional: a global `mushi` command

The wizard detects your framework and package manager, checks your credentials, installs the right SDK, writes framework-prefixed env vars (NEXT_PUBLIC_, VITE_, EXPO_PUBLIC_…) to .env.local, prints the snippet to paste, and offers to send a test report. It never overwrites an existing env var or edits your code.

Credentials come from the console: the project ID is on the Projects page; the API key is under Setup → Verify → Generate API key (report:write), not the AI keys in Settings. No project yet? The wizard can open the console for you.

For your editor, mushi setup writes the hosted MCP entry by default: the editor opens a browser sign-in on first use and no key is stored on disk. Add --stdio for a local server instead.

Commands

mushi --help and mushi <command> --help list every command and flag.

Common tasks

mushi doctor --fix                                  # check SDK, ingest and dispatch; fix what's safe locally
mushi reports list --status new --severity critical # what broke
mushi fix <reportId> --wait                         # an agent opens a draft PR for it
mushi fixes merge <fixId>                           # merge it and mark the report Fixed
mushi upgrade                                       # bump @mushi-mushi/* to the latest stable
mushi upgrade --check                               # CI: exit 0 current, 1 outdated, 2 registry down
mushi sourcemaps upload --release "$GITHUB_SHA" --dir ./dist
mushi ux ui                                         # the local UX studio (see @mushi-mushi/ux)

In CI or scripts, skip the prompts:

mushi init --yes --project-id <uuid> --api-key mushi_xxx --skip-test-report
MUSHI_API_KEY=mushi_xxx mushi connect --project-id <uuid> --endpoint <url> --wait   # wire a repo, wait for the first SDK heartbeat

Other init flags: --framework next, --cwd apps/web (monorepos), --endpoint <url> (self-hosted), --skip-install.

Environment variables

Config lives in ~/.config/mushi/config.json, written with mode 0o600 (a legacy ~/.mushirc is auto-migrated). --endpoint must be https:// except on localhost.

Programmatic use

Learn more

CLI reference · CLI and console loop · MCP setup

License

MIT