How do I add Pointer to my app?

Pointer mounts an in-browser feedback widget on your running application and connects your team's comment queue directly to your AI coding agents. This guide walks through prerequisites, the initialization command, interactive prompts, files written to disk, and non-interactive usage.

Prerequisites

• Node.js ≥ 18: Required to execute the initialization CLI.
• A Pointer account: An active account in your organization's workspace (see Workspaces).
• An API key: Found in the Pointer web dashboard. Click your user avatar in the top navigation, navigate to Profile, scroll to the API Key section, and copy the key (prefixed with ptr_). For details on storage and security, see API Keys.

The initialization command

Run the initialization command from the root directory of your project repository:

npx -y pointer-feedback init

Running without flags starts an interactive setup assistant that authenticates your session, links or creates a project, detects your dev server and frontend stack, configures widget embedding, and installs AI agent skills.

Walkthrough of interactive prompts

1. Pointer server URL: Defaults to https://api.pointer.moamen.work (or the value of POINTER_SERVER / existing .pointer/config.json). The CLI reaches out to the server's /api/branding endpoint to discover the server name and product identity.
2. API key prompt: You are asked to paste your API key (terminal input is hidden for security). The CLI exchanges this key with the server via POST /api/auth/login-with-key to verify permissions and prints your confirmed identity (e.g. ✔ Signed in as Jane Developer (Developer)). You have up to 3 attempts before the command aborts.
3. Project selection: The CLI retrieves your workspace's projects. You can select an existing project from the list or choose ＋ Create a new project…. When creating a project, you provide a project name; the CLI suggests a slugified key matching ^[a-z0-9-]+$ and creates it immediately.
4. Environment: Select the environment where this codebase runs (local, staging, or production; defaults to local).
5. Dev-server URL auto-detection (Step 4b): When configuring the local environment, the CLI inspects your repository to detect the dev-server address:
   • Vite: Scans vite.config.{js,ts,mjs,mts} for server.port and server.https, or checks package.json dev scripts for --port. Defaults to http://localhost:5173.
   • Angular: Scans angular.json under architect.serve.options.port and ssl, or package.json start scripts. Defaults to http://localhost:4200.
   • Next.js: Inspects package.json dev scripts for -p or --port, and checks .env / .env.local for PORT. Defaults to http://localhost:3000.
   • Create React App (CRA): Inspects package.json start scripts and .env for PORT. Defaults to http://localhost:3000.
   The CLI asks Where does this app run in local? with the detected URL prefilled as the default. Press Enter to accept it or enter your custom URL.
6. AI tool selection: The CLI checks environment variables (such as CLAUDECODE, ANTIGRAVITY_AGENT, TERM_PROGRAM for Cursor, WINDSURF, or OPENCODE) and asks you to confirm your AI coding environment (claude-code, cursor, windsurf, opencode, antigravity, or other).

What the command writes to your repository

Running init creates or updates the following files in your repository:

Deterministic widget injection & stack hand-off

How the feedback widget gets mounted into your application depends strictly on your frontend stack:

Deterministic injection (Vite & static HTML)

The CLI performs deterministic, automated widget injection for two stacks:

• Vite: Injects an environment-guarded script snippet into index.html reading from %VITE_POINTER_*% placeholders, and sets VITE_POINTER_ENABLED=true, VITE_POINTER_SERVER, VITE_POINTER_PROJECT, and VITE_POINTER_ENV in your .env file (plus stub entries in .env.example). This prevents hard-coding production URLs into local builds.
• Static HTML: Injects the <script src=".../pointer.js" defer></script> tag and <pointer-feedback ...> element directly before the closing </body> tag of your index.html (or the file passed to --html).

Hand-off for Next.js, Angular, CRA, and monorepos

Deterministic direct file injection is not supported for Next.js (App Router / Pages Router), Angular, Create React App, or monorepos because root layouts, server components, and multi-package trees vary too widely for rigid string substitution.

Instead of modifying your application code unpredictably, the CLI outputs a clear hand-off message:

ℹ <stack> detected — automatic injection isn't supported for this stack yet.
  The pointer-init skill was installed for <tool>. Run it and it will mount the widget for you:
    claude -> /pointer-init (or @pointer-init for cursor)
  Config is already saved in .pointer/config.json, so the skill won't ask for the key or project again.

Your configuration is already stored in .pointer/config.json. You simply invoke the pointer-init skill inside your AI editor, and the agent mounts the widget safely following your framework's idioms.

Non-interactive mode (CI and automated setup)

To run init without interactive prompts (for CI pipelines, dev container provisioning, or automated setup scripts), supply --yes (or -y). When --yes is used, --key is mandatory, and you must supply either --project <key> or --create <name>.

npx -y pointer-feedback init \
  --yes \
  --server https://api.pointer.moamen.work \
  --key "$POINTER_API_KEY" \
  --project my-web-app \
  --environment local \
  --tool claude-code

Adding --json outputs structured JSON containing setup details and automatically implies --yes.

CLI flags reference

Exit codes

Keeping the AI skills up to date

init copies the AI skills and pointer.sh into your repository, where they stay exactly as they were on the day you installed them. The server keeps changing. A skill file installed months ago is frozen prose describing an API that has moved on — and nothing about it looks wrong, which is what makes it expensive.

Every served skill and script carries a version stamp, so your installed copy can be compared with the server’s:

• npx -y pointer-feedback doctor reports a stale warning naming each file that is behind.
• npx -y pointer-feedback update --check lists what would change and writes nothing.
• npx -y pointer-feedback update refreshes them in place, preserving the .agents/ symlinks.

A stale skill is a warning, never an error — it still works, it is just behind. Nothing updates silently: refreshing is always something you ask for.

Self-hosting: the stamp defaults to the API’s build version, so every deploy moves it. Set Pointer__SkillVersion to invalidate installed copies after editing skill prose without shipping new code.

Troubleshooting & common pitfalls

• Invalid API key (Exit 3): Verify that the API key was copied completely from your profile page and has not been regenerated. If you regenerated your key in the web dashboard, the previous key was revoked immediately.
• This account cannot create projects (Exit 3): Stakeholder or client accounts are scoped to viewing comments and cannot mint new projects. An organization owner or administrator must create the project or invite you with developer privileges.
• Key already exists, choose another: Project keys must be unique within your organization. Provide a distinct slug or attach to the existing project with --project <key>.
• Comments blocked after deploy: If you enable origin enforcement on your project, ensure the deployed origin matches your configured environment URL. See Project Settings.