npx skills add ...
npx skills add revenuecat/ai-toolkit --skill revenuecat-cli
Drive RevenueCat from the terminal with the `rc` CLI, an alternative to the RevenueCat MCP server for humans, CI, and agents. Covers install, authentication, command discovery, and output conventions. Referenced by the other RevenueCat skills whenever they offer a CLI path.
npx skills add revenuecat/ai-toolkit --skill revenuecat-cli
rc is the official RevenueCat command line interface. It covers most of the same project, app, product, entitlement, offering, paywall, chart, and store-state operations as the RevenueCat MCP server (the CLI adds paywall generate/edit; the MCP server has some SDK feature-gate and experiment tools the CLI does not). Use whichever surface is available; this skill is the reference for the CLI path.
npx @revenuecat/cli <command> (good for CI and agent sandboxes).brew install RevenueCat/tap/rc, or npm install -g @revenuecat/cli.rc auth login (browser OAuth), or set RC_API_KEY. Non-interactively, pass --api-key or set RC_API_KEY.sk_...); that is correct because the CLI is a server-side tool. This is not the same as the public SDK keys (appl_/goog_/amzn_) you fetch for client app code. Never put a secret key in an app.The CLI is self-describing. Don't guess a command or flag, ask the binary:
rc commands --json lists the full command tree. Capability IDs appear in colon form (apps:create, products:store:plan).rc commands --schemas --json returns every command's flags, args, and examples in one call. This is the most reliable way for an agent to discover per-command detail.rc schema <command> --json returns detail for a single command, but the command must be space-separated tokens (rc schema apps create), not the colon form from rc commands. rc schema apps:create silently returns the root schema, so when in doubt use rc commands --schemas.rc --help and rc <noun> --help give human-readable help.--json gives stable, machine-readable output. Data goes to stdout, progress and chatter to stderr.{ "data": ... }, but the shape under data varies by command: lists are usually data.items[], though some commands use other keys (charts use data.names[]). Inspect the schema or the response rather than assuming data.items.--no-input never prompts, it fails instead. Use it in scripts, CI, and agents.--yes / -y skips confirmation prompts on mutating commands.--project-id <id> selects the project (or set a default with rc projects use). Pass it explicitly in scripts.0 success, 1 error, 2 bad usage, 4 auth, 5 not found, 6 rate limited.Look up exact flags with rc commands --schemas --json (or rc schema <space-separated command>). The common nouns:
rc projects list|create|use, rc apps list|create|keysrc products ..., rc entitlements ..., rc offerings ..., rc packages ...rc setup apple|google, rc products store plan|apply|syncrc charts list|show, rc customers ...rc paywalls generate|edit|publish (dashboard design/audit playbook: revenuecat-paywall-design)rc open [section] [id] --print (resource URLs: revenuecat-dashboard-links)Prefer the specific command. Drop to rc api <METHOD> <path> only for endpoints not yet in the CLI surface.