nightmarquee
# nightmarquee
Cinematic website prompts with live demos, inside your editor.
Art-directed prompts for hero sections, landing pages, SaaS, portfolio,
e-commerce and 3D/WebGL scenes. Each is tuned for Lovable, Bolt, Cursor,
Claude and Codex, and each has a live demo you can scroll.
NightMarquee is a remote MCP (Model Context Protocol) server. Add one address
to your AI tool, and search designs, get build prompts and starters, and check
a site from there:
```
https://mcp.nightmarquee.com/mcp
```
Some designs and tools ask you to sign in. Your tool opens the sign-in the
first time it needs it. The same steps, with buttons, are on
<https://nightmarquee.com/mcp>.
## Install
- **Claude**: Settings, then Connectors, add a custom connector, paste the address.
- **ChatGPT**: turn on developer mode in settings, add a new connector, paste the address. Developer mode is not offered on every account.
- **Cursor**: use the Add to Cursor button on <https://nightmarquee.com/mcp>, then confirm in Cursor.
- **VS Code**: use the Add to VS Code button on <https://nightmarquee.com/mcp>, check the server details, then install.
- **Claude Code**: `claude mcp add --transport http nightmarquee https://mcp.nightmarquee.com/mcp`, or as a plugin, which adds the remote server plus the skills below: `/plugin install nightmarquee --marketplace nightmarquee/nightmarquee-mcp`
- **Codex**: `codex mcp add nightmarquee --url https://mcp.nightmarquee.com/mcp`
- **Any other tool**: paste the address where it asks for a remote MCP server. It uses the streamable HTTP transport and signs you in with OAuth.
## The older way: this npm package
This package is the earlier way: a local stdio server (Node 20 or newer) with
the tools listed under Use below. It still works. Pick it if your tool cannot
add a remote address.
**Claude Code**
```bash
claude mcp add nightmarquee -- npx -y nightmarquee
```
**Claude Desktop / Cursor**: add to your MCP config:
```json
{
"mcpServers": {
"nightmarquee": {
"command": "npx",
"args": ["-y", "nightmarquee"]
}
}
}
```
Claude Desktop reads `~/Library/Application Support/Claude/claude_desktop_config.json`
on macOS and `%APPDATA%\Claude\claude_desktop_config.json` on Windows. Cursor
reads `~/.cursor/mcp.json`, or `.cursor/mcp.json` inside a project.
**Codex**: the CLI, the IDE extension and the ChatGPT app share one config.
```bash
codex mcp add nightmarquee -- npx -y nightmarquee
```
That writes this to `~/.codex/config.toml`, which you can also add by hand
(or to `.codex/config.toml` in a trusted project):
```toml
[mcp_servers.nightmarquee]
command = "npx"
args = ["-y", "nightmarquee"]
```
Codex uses the tools but does not show MCP prompts as commands yet, so ask for
a design in plain language and it calls `nightmarquee_search_prompts`. Prompts
come back in the Codex dialect unless you pass another `tool`.
No account needed. Free prompts work immediately.
## Use
- `nightmarquee_search_prompts`: browse by `query`, `category` or `tier`
- `nightmarquee_get_prompt`: fetch one by `slug`, optionally with your own
`brand` name and a `tool` dialect (`lovable`, `bolt`, `cursor`, `claude`,
`codex`). Without one you get `codex` in Codex and `claude` elsewhere
- `nightmarquee_get_footage`: fresh download links for a demo's footage, when
the ones from `nightmarquee_get_prompt` have expired
- `nightmarquee_sign_in`: connect an account, so the prompts that come with a plan open
- `nightmarquee_whoami`: check what's connected
Categories are `hero`, `landing`, `saas`, `portfolio`, `ecommerce`, `3d`, `mobile` and `dashboard`.
Ask for a design in plain language and the model will usually reach for
`nightmarquee_search_prompts` on its own. Or drive it directly:
```
nightmarquee_get_prompt(slug: "glasshouse-nine", tool: "cursor", brand: "Northwind")
```
Every prompt in the catalog is also registered as an MCP prompt, so the whole
library shows up in your client's command palette. In Claude Code that means a
slash command per design:
```
/mcp__nightmarquee__glasshouse-nine
```
Other clients surface MCP prompts their own way. Claude Desktop lists them in
the attachment (`+`) menu; Codex does not list them yet. The list is built from the catalog fetched at
startup, so if the server can't reach nightmarquee.com when your editor
launches, the tools still work but the per-prompt commands won't appear until
you restart.
## Footage
Most demos play video. When your account may download it, `nightmarquee_get_prompt`
returns the footage from the demo, every clip, with a signed link per file, and
the prompt text ends with a note naming each clip. The agent downloads the files
into `public/` under their names and then builds, so the page you get plays the
same footage as the demo. The links last 10 minutes; `nightmarquee_get_footage`
fetches fresh ones without the prompt.
A free prompt's footage needs a free account, and the rest come with any
plan. Without an account, the prompt text still arrives in full, with one
line saying how many clips the demo plays and that `nightmarquee_sign_in` gets
them. Each account can fetch 30 packs an hour, shared with downloads on the
site.
In Lovable and Bolt, download the files from the prompt page on
nightmarquee.com and upload them there.
## Skills
The same catalog also installs as agent skills, for editors that read
`SKILL.md` files:
```
npx skills add nightmarquee/nightmarquee-mcp
```
That adds a `nightmarquee` guide and one skill per free prompt. They carry no
prompt text: each fetches the current prompt through the remote server,
`https://mcp.nightmarquee.com/mcp`, when it is connected, and otherwise asks
you to copy it from the design's page on nightmarquee.com.
## Plans
Some prompts are free, currently `glasshouse-nine`, `aetheris-voyage`,
`nordlys-fintech`, `solstice-festival`, `lumen-observatory`, `sumi-atelier`,
`wren` and `fieldnote`. The rest come with
[any plan](https://nightmarquee.com/pricing?ref=mcp). Run
`nightmarquee_search_prompts` with `tier: "free"` for the live list.
Run `nightmarquee_sign_in` and approve the code in your browser. It waits about
a minute; if you took longer, run it again and it picks up the same request
rather than issuing a new code. Either way the rest unlock in the same session,
no restart. Fetching a locked prompt without an entitlement returns a
teaser and a pricing link, never the body.
For headless environments where no browser is reachable, carry a token across
instead: sign in once on a machine that has a browser, copy the `token` value
(`nm_live_…`) out of `~/.nightmarquee/credentials.json`, and set it there.
```json
{ "env": { "NIGHTMARQUEE_TOKEN": "nm_live_..." } }
```
Connections are listed on [your account page](https://nightmarquee.com/account)
and can be revoked from there; the next call falls back to free-only.
## Configuration
| Variable | Effect |
|---|---|
| `NIGHTMARQUEE_TOKEN` | Use this token instead of the stored one. Wins over `credentials.json`. |
| `NIGHTMARQUEE_API` | Point at a different backend. Defaults to `https://nightmarquee.com`. |
| `NIGHTMARQUEE_HOME` | Directory for `credentials.json`. Defaults to `~/.nightmarquee`. |
| `NIGHTMARQUEE_TIMEOUT_MS` | Ceiling on every API request, in milliseconds. Defaults to `10000`; unparseable or non-positive values fall back to the default, and anything above `2147483647` is clamped to it. |
| `NIGHTMARQUEE_NO_TELEMETRY` | Set to `1` (any non-empty value works) to stop sending the install ID and your client's name. See below for exactly what this does and doesn't cover. |
## Network calls
The remote server runs on our side and this package is not involved. When you
run the package, it makes HTTPS requests to `nightmarquee.com` (or the address
in `NIGHTMARQUEE_API`), and to no other host:
- `https://nightmarquee.com/api/mcp/*`: the catalog, the prompt text and the
sign-in.
- `https://nightmarquee.com/api/footage/<slug>`: the footage links for a demo.
Each request carries the headers listed under Privacy Policy.
## Privacy Policy
Full policy: <https://nightmarquee.com/privacy>
### What is collected
Credentials are stored locally at `~/.nightmarquee/credentials.json`, mode 0600.
Fetching a prompt is an HTTP request to nightmarquee.com, so the server
necessarily sees which prompt slug and tool dialect you asked for, and logs
that. That is true in every configuration.
On top of that, each request carries three headers by default:
- `x-nm-install`: a random UUID generated on your machine on first run and
kept in `credentials.json`. Not derived from anything, tied to nothing else.
- `x-nm-client`: the name and version your MCP client reports, e.g.
`claude-code/2.1.240`. In Codex this package also asks for the Codex dialect
itself, so turning telemetry off does not change which prompt text you get.
- `x-nm-version`: the version of this package.
If you are signed in, the request also carries a bearer token identifying your
NightMarquee account.
**Never sent, in any configuration:** the prompts or code you write, file
paths, file contents, or anything else from your project.
### How it is used and stored
Request records are used to operate the service and to understand which prompts
people use, nothing else. They are held in our Supabase database.
### Opting out
`NIGHTMARQUEE_NO_TELEMETRY=1` drops `x-nm-install` and `x-nm-client`.
**`x-nm-version` is still sent**, and requests are still made and still logged
server-side, and the log rows just carry no install ID and no client name. If you
are signed in, the bearer token identifies your account regardless of this
setting.
### Third-party sharing
Your data is not sold, rented, or shared with third parties for their own
purposes. It is processed only by the infrastructure providers running the
service on our behalf: Vercel (hosting), Supabase (database), and Lemon Squeezy
(payments, for subscribers only). No advertising or analytics networks receive
any of it.
### Retention
Account data is kept while your account exists. Email
<support@nightmarquee.com> to delete your account and we remove your data
within 30 days, except records we must keep for tax and accounting (held by the
merchant of record).
Request records are not currently expired on a fixed schedule; they are deleted
on request along with your account.
Tokens and local credentials are under your control at any time: revoke a
connected editor from your [account page](https://nightmarquee.com/account),
which stops the token working at once (the record is kept, marked revoked), or delete `~/.nightmarquee/credentials.json`
to remove the local copy.
### Contact
Privacy questions, data access, and deletion requests:
<support@nightmarquee.com>
## License
The server code in this package is MIT.
The prompt text it fetches is licensed separately: free prompts cover personal
and non-commercial projects, a plan adds commercial use, and the prompt text
itself may not be republished or resold. Full terms:
<https://nightmarquee.com/license>.
MIT © NightMarquee
TDQS
Scored across 5 tools
Each tool has a clearly distinct role: identity (whoami), catalog search (search_prompts), prompt retrieval (get_prompt), footage links (get_footage), and authentication (sign_in). The only surface overlap is that get_prompt also bundles footage links, but its description explicitly frames get_footage as the fallback for expired links, so boundaries are clear.
All names use a consistent snake_case pattern with a shared 'nightmarquee_' prefix and verb-based actions (get_, search_, sign_). 'whoami' is the sole stylistic outlier but is a widely recognized convention for an identity probe.
Five tools is well-scoped for a prompt-catalog consumer: search, retrieve prompt, retrieve footage, authenticate, and check account status. Each tool earns its place with no redundancy or padding.
The surface covers the full consumption lifecycle from catalog search through prompt and footage retrieval, plus auth/identity, with get_footage handling link-expiry recovery. Minor gaps exist around explicit category/tier browsing or prompt-list enumeration, but search covers most of that ground.