Your Hookie account, from the command line.
Everything the console and the MCP tools can do, as commands: manage projects, endpoints, destinations and workflows, search what arrived, and redeliver what failed. Built for scripts, CI jobs and AI agents as much as for people, with --json on every command and a real exit code when something goes wrong.
npm i -g @hookieai/cli
hookie login
hookie projects listAuthenticates with the same OAuth connection a coding agent uses — read-only until you widen it, revocable in one click.
The commands you will reach for
hookie endpoints create --name Checkout --slug checkout --dataset ordersSet up from a script
Projects, endpoints, destinations, rules and workflows: create and change them from a terminal, a CI job or an agent, with the same validation the console applies.
hookie deliveries search --status failed --since 24hFind what failed
Search what your account stored (events, raw submissions, deliveries) with a time window and paging. --json gives one document an agent or jq can act on.
hookie replay --since 1h --failedRedeliver a bad hour
One delivery, or every failure in a window: each event once per destination, never one still being retried, and it stops rather than piling on when a destination pushes back.
hookie listen --forward http://localhost:3000/hooksDevelop against live events
Forwards your project's live events to a server on your machine and prints each one's local status and latency. Your server's answer never touches the real delivery, it reconnects and resumes without sending an event twice, and the endpoint's URL never leaves Hookie. The live stream is best-effort, so a long disconnect is reported as a gap.
hookie send --endpoint checkout --data @event.jsonSend a test event
Posts to the endpoint's public URL like a real sender, so it takes the same path through ingest, routing and delivery. The answer says where it was routed.
Made for AI agents as well as people
- Every command goes through your account in the app. The CLI changes and reads what the console shows. The one thing it brings to your machine is the live event stream, for
hookie listen, and only to a URL you name. - One connection, the same as an MCP client's. It starts read-only, you widen it in the console, and you can revoke it there in one click.
- Safe to automate. A destructive command asks first, and with no one to ask (a script, an agent) it refuses unless told
--yes. Exit codes say whether to fix the input, reconnect, or retry.
In your code, not just your terminal
@hookieai/node is a zero-dependency SDK for servers, Cloudflare Workers, Deno and Bun. Send events with idempotency keys, verify the signature on every webhook Hookie delivers to you (Hookie-Signature and Standard Webhooks), and manage endpoints, destinations and deliveries from code. @hookieai/forms does the same for HTML forms in the browser.
Everything the console does, scriptable
Forty commands, generated from the same catalog that defines the tools a connected AI agent gets over MCP, so the CLI, the console and your agent are the same API, and a new capability appears in all three at once.
- List and inspect
hookie endpoints list --json | jq '.webhooks[].base_slug'- Create from a script
hookie endpoints create --name Checkout --slug checkout --dataset orders- Search deliveries
hookie deliveries search --status failed --since 24h- Query a dataset
hookie datasets query --dataset orders- Open the console here
hookie open --project growth --tab deliveries
hookie <command> --help (or hookie help <command>) prints its usage, every flag with its type, and an example; for a generated command they come from the schema. An unknown flag is an error with a suggestion, never a request that silently did something else.
Your workspace, in a file you can review
hookie diff shows what would change; hookie apply makes it so, after asking. Endpoints, rules and destinations live in one file you can put in a repository and change in a pull request.
It is honest about the limits rather than pretending to be a reconciler: a difference the admin API cannot change in place is reported with the reason, and nothing the file omits is ever deleted unless you ask for it with --prune.
# hookie.yml — check it in, review it in a PR
project: default
endpoints:
- name: Checkout
slug: checkout
dataset: orders
criteria: [{"path": "type", "equals": "order.created"}]
destinations:
- name: Warehouse
url: https://warehouse.internal/hook
dataset_filter: ordersStart with one command
Free tier included. Sign in once, and the CLI manages the same account as the console.