Scoopkit has an official command line tool, published to npm as scoopkit. It runs without installing anything:
`
npx scoopkit taxonomy
npx scoopkit events list --category MODEL_RELEASE --limit 3
`
There are five commands (login, whoami, events, search, taxonomy), one runtime dependency, and a --json flag on every one of them. Reading events needs a free key; search needs Pro. The whole thing is a thin wrapper over the REST API, which is deliberate, because the useful thing about a CLI here is not new capability but the absence of setup.
Verified 2026-09-05 by running npx scoopkit@latest fresh from the public npm registry against the live production API, not from a local checkout. Version at time of writing: 0.1.0.
The five commands
| Command | What it does | Tier |
|---|---|---|
login | Emails a magic link, waits for the click, saves the key locally | Free |
whoami | Shows tier, email, and remaining quota for the saved key | Free |
events list | Lists events, newest first, with filters and cursor pagination | Free |
events get <id> | Fetches one event by ID | Free |
search <query> | Semantic search over events | Pro |
taxonomy | Prints all 10 categories and 57 subcategories | Free, no key needed |
events list takes --category, --subcategory, --since, --until, --limit, and --cursor. Those map one to one onto the REST query parameters, so anything you learn here transfers directly to the HTTP API and back again.
taxonomy needs no key at all, which makes it the fastest way to see whether the tool works before committing to anything:
`
$ npx scoopkit taxonomy
MODEL_RELEASE (free) Model Release & Capability
MODEL_RELEASE.NEW_MODEL_RELEASE (free)
MODEL_RELEASE.MODEL_VERSION_UPDATE (free)
MODEL_RELEASE.OPEN_WEIGHTS_RELEASE (free)
...
`
Why npx rather than pip or brew
The install step is where most people quit. npx scoopkit taxonomy has no install step, no virtualenv, no PATH question, and no decision about whether to install globally. Someone evaluating whether this API is worth a key can see real data in one line, and if the answer is no, nothing was installed to clean up.
The second reason is agents. When a coding agent needs AI industry news on someone's behalf, it shells out. An npx one-liner is a shape agents already handle well, with no environment preparation and no state left behind. That is the same reason the tool prints a clean table by default and full JSON under --json: a human reads the table, an agent pipes the JSON.
Scoopkit's API is written in Python, so PyPI was the obvious first instinct. It lost on exactly this point. pipx install and uvx are fine, but they are not what the agent tooling ecosystem reaches for first. PyPI stays on the list as a secondary target rather than the priority.
Auth, and the part that surprises people
npx scoopkit login asks for your email, sends a magic link, and waits. You click the link in your inbox, confirm on the page that opens, and the key lands in ~/.config/scoopkit/config.json. SCOOPKIT_API_KEY in the environment overrides the stored key, which is what you want in CI or a container where writing a config file is pointless.
The confirmation click is a real button, not just opening the link. That design exists because corporate mail scanners prefetch every URL in an incoming email to check it for malware. When the verification link consumed the single-use token on GET, the scanner burned the token seconds after send and the human never got their key. Splitting the read-only page from the token-minting POST fixed it. It is a good example of the kind of bug that only shows up against real email infrastructure, and it would have silently blocked signup for a large share of the developers this API is for.
A bug the CLI found in the API
Building the tool caught a real defect. /v1/auth/me returns an envelope, {data: {...}, meta: {...}}, like every other endpoint. The first version of whoami read tier and email off the top level of the response instead of out of data, so it cheerfully printed tier: undefined and email: (none) while the API was returning correct information the whole time.
Nothing was wrong on the server. The client was wrong in a way that looked exactly like the server being wrong, which is the failure mode a first-party client is best positioned to catch. Writing a consumer of your own API is the cheapest integration test there is.
FAQ
Do I need an API key to try it?
Not for taxonomy. Everything else needs a key, and the free tier gives you 100 requests a day with a 45-day archive, usable in production. Get one with npx scoopkit login or from the quickstart.
Does it work without Node installed globally?
It needs Node 18 or newer, because it uses the built-in fetch. It has one runtime dependency (commander) and no native modules, so there is nothing to compile.
Is this different from the MCP server?
Yes, and they are for different things. The CLI is for shelling out from a terminal or a script. The MCP server is for an AI assistant that speaks Model Context Protocol and wants tools rather than a command line. Same API underneath, same key.
What does --json change?
It prints the raw API response instead of the formatted table, including the pagination cursor. That is the flag to use when something downstream is parsing the output.