CLI · ontrackd docs

The package installs two commands: feedback, for setting up, linking, pulling and plain sites, and feedback-mcp, the server your agent talks to. Run them through your package manager, from the app the package is installed in:

$ npx feedback --help

$ pnpm exec feedback --help

$ yarn feedback --help

$ bunx feedback --help

feedback --help prints the list below in short. link, pull and push work on the app in the current directory, or on the folder you pass them.

Set up

init is the command you run before the package is installed, so it goes through npx, in your app’s directory:

npx @ontrackd/feedback init

Command What it does

feedback init Set ontrackd up in this app: detect the framework and the package manager, install the package as a dev dependency, add the plugin to the framework’s config, and register the MCP server with your agents. It first asks who will comment, then lists what it changed and what to do next. In an agent’s shell, where it cannot ask, it prints the question and the two commands that answer it, and changes nothing. A config it cannot edit safely is left alone: it prints the file and the lines to add by hand.

Command	What it does
`feedback init`	Set ontrackd up in this app: detect the framework and the package manager, install the package as a dev dependency, add the plugin to the framework’s config, and register the MCP server with your agents. It first asks who will comment, then lists what it changed and what to do next. In an agent’s shell, where it cannot ask, it prints the question and the two commands that answer it, and changes nothing. A config it cannot edit safely is left alone: it prints the file and the lines to add by hand.

Flags of feedback init:

Flag	What it does
`--just-me`	Only you comment, on your dev server. No account needed.
`--with-previews`	Others comment too, on shared previews, with your ontrackd account. After setting the app up, `init` connects it to a project: it prints a link and a code (and opens the link when it can), you check the code, pick or create the project in the dashboard and approve, and `init` writes `FEEDBACK_SECRET` to the app’s `.env.local` (git-ignored). It waits up to 9 minutes; run the same command again to keep waiting on the same code, which lasts 10 minutes. An app that already has a secret is not connected again.
`--reconnect`	With `--with-previews`: connect again, to the project you approve, even if the app already has a secret.
`--app <dir>`	The app to set up, in a monorepo. Defaults to the current directory; at a workspace root, `init` lists the apps it found.
`--framework <id>`	Skip detection: `vite`, `vue`, `svelte`, `sveltekit`, `preact`, `solid`, `react-router`, `tanstack-start`, `rsbuild`, `astro`, `nuxt`, `next`, `webpack` or `rspack`.
`--dry-run`	Show what would change, and change nothing: no install, no file written.
`--yes`, `-y`	Write without asking for confirmation.
`--no-mcp`	Do not register the MCP server with your agents.

Link and pull

Command	What it does
`feedback link`	Sign in with an emailed code, choose a project, and save its secret in `.feedback/credentials.json`, which git ignores: the same as putting `FEEDBACK_SECRET` in `.env.local`. Then import the comments you left locally, after a report and a confirmation.
`feedback pull`	For working without an agent: fetch the comments left on previews, each pointing at its `file:line` in your code as it is now. Your dev server’s panel then includes them. An agent connected over MCP fetches them itself.
`feedback push`	Upload what a preview build could not send, so its comments point at their lines. The build says when to run it.

Flags of feedback link:

Flag	What it does
`--project prj_…`	Skip the project prompt. You must be the project’s owner or an admin.
`--email <address>`	Sign in with this address instead of asking for it.
`--dry-run`	Show the import report and import nothing.
`--yes`, `-y`	Import without asking for confirmation.
`--no-import`	Only link; leave local comments where they are.

Cloud mode

For syncing comments you write on your dev server to a Cloud project. To import the ones you already left once, feedback link does it too.

Command	What it does
`feedback remote login`	Sign in to the Cloud API given with `--api <url>`, with an emailed 6-digit code. The token is kept outside the project, in `~/.config/feedback/`.
`feedback remote init --cloud`	Link this app to a Cloud project: pick one, or create one with `--new`. Restart the dev server afterwards.
`feedback remote whoami`	Show which account you are signed in as.
`feedback remote logout`	Revoke the sign-in token and forget it on this machine.
`feedback remote status`	Show the API, project, session and sync state of this app.
`feedback remote source-excerpts`	`on` or `off`: whether Cloud comments include the source lines around the element. Restart the dev server afterwards.
`feedback invite`	Create a review link for a reviewer: their name, then `--preview <url>`. Works through the running dev server. `invite list` lists them; `invite revoke <participant>` cuts one off.

Flags of feedback remote login:

Flag	What it does
`--api <url>`	The API to sign in to, like `https://api.ontrackd.com`. Defaults to `FEEDBACK_API_URL`, then to the API this app is linked to.
`--email <address>`	Sign in with this address instead of asking for it.
`--name "<name>"`	The display name reviewers see.

Flags of feedback remote init --cloud:

Flag	What it does
`--project prj_…`	Skip the project prompt.
`--new "<name>"`	Create a project with this name and link it.
`--api <url>`	The API to use, when you are signed in to more than one.
`--no-source-excerpts`	Send only the file and line of each comment, not the source lines around it.
`--replace`	Replace an existing link. The old one moves to `.feedback/backup-<time>/`; keep it.
`--port <port>`	The sidecar’s port, when it is not the default.
`--root <path>`	The app to link, when it is not the current directory.

Plain sites and Hugo

Command	What it does
`feedback serve`	Serve a folder of HTML files for review, each comment pointing at its file and line. `--port`, `-p` (default `5500`); `--host` (default `localhost`).
`feedback build`	Copy a folder of HTML files for deployment. With `FEEDBACK_SECRET` set it is a shared preview; without it, every file is copied unchanged. `--out`, `-o`: the output folder. `--clean`: empty it first.
`feedback hugo server`	Run `hugo server` for review, each comment pointing at its file and line. Takes Hugo’s own flags.
`feedback hugo`	Build with Hugo. A shared preview when `FEEDBACK_SECRET` is set, otherwise plain `hugo`.

MCP server

Command	What it does
`feedback-mcp init`	Print the exact command for your agent, for this app or every app in the monorepo.
`feedback-mcp`	The MCP server itself, as `init` prints it. Repeat `--root` for several apps.

Flags of feedback-mcp:

Flag	What it does
`--root <app>`	An app to serve. Repeatable; defaults to the current directory.
`--port <port>`	The sidecar port to look for first (default `7331`).

The tools it gives your agent are in MCP tools.

Environment variables

Variable	What it does
`FEEDBACK_SECRET`	The project’s secret. On your host, set it for preview builds only, where it turns shared mode on; a build without it carries nothing of ours. On your machine, put it in the app’s `.env.local`: the CLI and the MCP server read the environment first, then `.env.local`, then `.env`, then what `feedback link` saved.
`FEEDBACK_API_URL`	The API to use, over the one saved by `feedback link`. Defaults to `https://api.ontrackd.com`.
`FEEDBACK_CLOUD_SYNC`	Set to `0` to keep a linked app local: its dev-server comments are not synced, and your agent does not fetch the comments left on previews.