Quickstart
Install crab, build and call a service locally without an account, then deploy to Cloud when ready.
1. Install the CLI
curl -fsSL https://crabtalk.ai/install.sh | shcurl -fsSL https://crabtalk.ai/install.sh | shirm https://crabtalk.ai/install.ps1 | iex- It installs the latest release, crab 0.1.6, to
~/.local/bin, or toCRAB_INSTALL_DIR, after checking the download against the SHA-256 GitHub records for it. On Windows it adds that directory to your userPATHwhen it is not on it. - Releases are built for macOS and Linux on x86_64 and arm64, and for Windows on x86_64.
crab upgradereplacescrabwith the latest release when it is newer.
2. Create and build a service
crab new notes
cd notes
crab buildcrab new writes a service with echo and durable remember/recall tools. crab build installs any missing build tools and builds its image. No account or login is needed for the local loop.
3. Run it locally
crab dev -d
crab --context default deploy
crab --context default status
crab --context default inspect notes
crab --context default call notes echo '{"query":"hello"}'
crab --context default call notes remember '{"message":"my first note"}'
crab --context default call notes recall '{}'crab dev -d starts the host in the background at http://127.0.0.1:7777. crab deploy builds and deploys to it; run it again after changing the service. --context default selects the local host even if you have signed in to Cloud before.
crab inspect prints the service's tools and their arguments. The call returns {"query":"hello"}, with usage on stderr. recall returns "my first note" from persistent storage, including after a host restart. The starter shares this message across callers; see Data for per-user storage. Your agents can call the same tools over local MCP. Stop the host with crab dev stop.
4. Deploy to Cloud
When you want the service on CrabTalk Cloud, sign in:
crab loginIt opens a browser to sign in with Google, or GitHub with --github, and switches to the cloud context. On a machine with no browser, --device prints a link and a code to approve from another one.
On Hobby, Cloud deployment requires GitHub verification first: crab billing verify (E0001). Verification grants $1 in one-time credit valid for 30 days; after it ends, choose a paid plan to continue hosting. Data is kept for 30 days after credit ends, then deleted. These requirements apply to Cloud hosting; local development stays account-free.
crab --context cloud deployIt builds and deploys the same service to your account. It prints a link last: open it, review what it deploys, and approve. On Cloud, services are addressed as owner/name: use crab inspect owner/name to read their tools and crab call owner/name <tool> '<arguments as JSON>' to call one.
Writing its tools: Writing a service. Running it on your machine first: Local host.