HookovaHookova MCP server: setup guide
Your save library, as tools an assistant can call. Install it once and “save this page,” “what did I save about sourdough,” “tag those three as Recipes” and “summarise everything I saved this week” all become things you can just say. It takes about two minutes and needs no account beyond the one you already have.
Or don’t do any of it yourself: tell an assistant with a terminal install www.hookova.com/install/mcp and it fetches the commands and runs them. What that does.
What it is
MCP — the Model Context Protocol — is the standard way an AI client asks an outside system to do something. An MCP server is a small program that advertises a set of tools; the client reads that list and calls them when a conversation needs them. The Hookova server advertises forty-five: four to get itself signed in, six pointed at your save library, two at transcription, two at your projects, one that makes images, two that make promotion videos, eleven at publishing to your channels, thirteen at the free sound-effect and music libraries, and four at the free color LUTs.
It comes two ways. The hosted server is a URL — https://www.hookova.com/mcp— that you add to your AI app as a connector; it signs in through your browser and needs nothing installed (next section). The local server is described in the rest of this page.
Two consequences worth knowing before you install the local one. First, it is local: nothing is hosted, there is no Hookova process watching your conversations, and the only thing that leaves your machine is the API call a tool makes. Second, it is your account: one token, one library, no sharing — and saves made here count against the same daily allowance as everywhere else.
It works with any client that speaks MCP over stdio. The instructions below cover Claude Code, the Claude desktop app and Cursor; anything else follows the same shape.
Connect with a URL
If your AI app can add a remote MCP server — Claude (the app, claude.ai and Claude Code), ChatGPT, Cursor, most current clients — this is the whole setup. No zip, no Bun, no token to copy.
https://www.hookova.com/mcpIn the Claude app or on claude.ai: Settings → Connectors → Add custom connector, and paste it. In Claude Code:
claude mcp add --transport http hookova https://www.hookova.com/mcp --scope userthen run /mcp, pick hookovaand choose Authenticate. Either way the app opens Hookova in your own browser; you sign in there — Google works, because it’s your browser and not an agent’s — check the account, and press Allow.
What the app gets. A key that works for an hour and is renewed from a connection you can see and end: Settings → MCP serverlists every connected app with a Disconnect button. The key reaches the same tools as the local server’s and nothing else — it can’t open the website as you. The hosted server has thirty-six of the forty-five tools — not the sign-in tools (the app signs itself in), not the four sound and music downloads (it has no disk; each result carries the MP3’s link), and not download_lut or apply_lut (they grade files on your machine) — plus create_upload. Its publish_posttakes the file from the chat instead of your disk — a file you attach in ChatGPT, or a public link to the file, up to 100 MB a post — and where the chat cannot pass a file (Claude), create_upload hands you a hookova.com page to pick it from your device. So you can create, read, edit, schedule and cancel posts from the conversation.
What can Hookova do?
Find references, build from an inspiration project, then publish your finished video.
Get Inspired
Browse My Saves, search references, and organize tags and notes. Analyze Video in the app creates shot-by-shot breakdowns; MCP can read existing analysis context but cannot start an analysis.
MCP tools: save_link, list_saves, get_save, update_save, delete_save, export_saves
Make Videos
Start with Inspiration Projects: read a board and export its hook guide for your AI video workflow. Make the stills with generate_image — Pop-Cam's image model, 1 credit per image, credits provided by Pop-Cam. Or have the whole video made: make_promo_video turns a product brief into a finished 30-40 second promotion video in about 20 minutes, for 10 credits refunded if it fails, and list_promo_videos checks on it. Then use Publish to upload, draft, schedule, or post finished media to connected accounts. Apart from that promotion video, the Hookova MCP server does not render videos.
MCP tools: list_projects, get_hook_guide, generate_image, make_promo_video, list_promo_videos, list_publish_accounts, list_publish_groups, create_publish_group, update_publish_group, delete_publish_group, publish_post, list_posts, get_post, update_post, schedule_draft, cancel_post
Tools & Assets
Transcribe Video: `transcribe` takes a saved clip or a link and `get_transcription` reads the result (credits, on success only). Sound Effect Library and Music Library: search, browse and download free MP3s — no sign-in, except `recommend_music`, which ranks tracks for a described video. On the hosted server the results carry the MP3 url instead of downloading. Color LUTs: list_luts and get_lut select a free sRGB creative look; the local server also downloads cubes and applies filters with FFmpeg. Subtitle Library is coming soon.
MCP tools: list_luts, get_lut, download_lut, apply_lut, transcribe, get_transcription, search_sounds, list_sounds, list_sound_groups, get_sound, download_sound, download_sounds, search_music, list_music, list_music_tags, recommend_music, get_track, download_track, download_tracks
Analyze
Coming soon. This section is separate from Analyze Video under Get Inspired; no analytics MCP tools are available yet.
Connect to My AI
Open Connect to My AI to set up your assistant. Prefer browser sign-in with connect, then check auth_status. sign_out forgets the local token; it does not revoke it.
MCP tools: connect, auth_status, sign_in, sign_out
Ask your assistant: “Find my saved references”, “Get the hook guide for my inspiration project”, or “Prepare a draft for my connected accounts”. Approve the accounts, media, caption, and timing before asking it to publish.
The forty-five tools
You never call these by name — you ask for what you want and the client picks. They are listed here so you know what it’s allowed to do, which is the useful thing to know about any tool you hand an assistant. Four cover signing in; six cover the library; two cover transcription; two cover projects; one makes images; two make promotion videos; eleven cover publishing; thirteen cover the sound-effect and music libraries; four cover color LUTs.
Signing in
These exist so a token never has to arrive by config edit and restart. If you install the server with no token at all, everything else answers “Sign in to save.” and connectis the way out of it — from inside the conversation, in one turn, without a credential ever being typed into the chat.
Signs this server in through your browser — the way every other command-line tool does it.
Returns a link; you open it, approve the account you're already signed into, and the page hands the key back to this machine over 127.0.0.1. Call connect again to wait for the approval. The link lasts five minutes and works once. Use Chrome or Edge — Safari won't let a web page reach your own machine. It supports Google accounts, and the key never appears in the conversation.
Whether there's a working token, whose account it is, and what that account can do today.
Plan, credits left, saves used against the daily allowance — and where the token came from: the environment variable, a saved sign-in, or this session. Call it first whenever anything answers “Sign in to save.” It changes nothing and never returns the token.
Signs this server in without editing config or restarting the client.
Pass a token — minted or a session cookie — or an email and password. The token is verified before it's kept, saved at mode 0600 so the next run starts signed in (remember: false keeps it to the session), and never echoed back.
Forgets the token here and deletes the saved file.
Not revocation: the token stays valid until its 30 days are up, because there is no endpoint to revoke one. If HOOKOVA_TOKEN is in your config it takes over again at the next restart, and the answer says so.
Prefer connect to all of this. Nothing is typed, nothing is pasted, and no credential enters the chat log at all. Failing that, prefer the token argument to a password: a password passed as a tool argument is written into the conversation transcript wherever your client keeps it, and a Google-linked account has no password to pass anyway. HOOKOVA_EMAIL/HOOKOVA_PASSWORD in the config do the same job without putting a credential in a chat log.
The library
Saves any public URL — a TikTok, Reel or Short, or an ordinary web page.
Returns immediately with status queued; the title, synopsis, takeaways and category are written by a background worker a few seconds later. Saving the same URL twice returns the existing save rather than a duplicate.
Browses and searches the library, newest first.
q searches title, synopsis, URL, notes and tags; tag filters to one category. Also returns every tag with a count and the day's remaining save allowance. Page with the returned nextCursor.
One save in full: synopsis, excerpt, takeaways, tags, notes.
Also reports whether a shot-by-shot analysis exists for the same URL. Read this before update_save, because tags are written as a whole list.
Edits the fields you own on a save.
tags REPLACES the whole list — send the existing tags plus your addition, or the rest are deleted. Everything else on a save is the worker's and can't be set here.
Removes a save and its stored images.
Permanent. There is no undo and no trash to recover it from.
The library in bulk, newest first, with notes and takeaways.
For summarising or reorganising across many saves at once. Capped at 100 by default because the full export runs to thousands of rows — use list_saves to search instead.
Transcription
Starts a transcription — the timed spoken words — of a TikTok, YouTube, Reels or RedNote video.
A pasted link (nothing is saved) or a video already in your library. Spends credits, only if it succeeds: 1 for a saved clip, 1 per started 3 minutes for a link, up to 10 minutes. The assistant is told to say the price and wait for your yes. Asking twice never charges twice.
A transcription's progress, then its transcript as timed lines.
Queued, running (with a percentage), done or failed (and uncharged). Without an id, your recent transcriptions. text: true adds the whole thing as one paragraph.
Projects and the hook guide
A project is a board for one video in production. These two read it — and the second turns it into the material HyperFrames builds your own opening from. See From a board to a hook.
Every board you own: id, name, the brief in your words, how many saves are pinned.
A project is a board for one video in production. This is how a board is found by name; get_hook_guide takes the id. Cover images are dropped.
A board as HOOK-GUIDE.md — the material for building the opening of your own video.
The brief, every pinned save (synopsis, takeaways, a transcript excerpt), and for each pinned clip with a finished shot-by-shot analysis its hook pattern: the opening shot by shot — timing, shot size, camera, where the text lands, the technique each shot performs — plus reference stills. Pass dir and the server writes the file and downloads the stills into dir/refs/ itself. Free: it re-shapes analyses you already paid for and generates nothing. Nothing from the source clips is reused; the guide says so in its own text.
Images
The stills for a video — a character in every scene, a product shot, a thumbnail — made with Pop-Cam’s image model and paid from your Hookova credits.
One image from Pop-Cam's image model (Nano Banana 2.1): from a prompt, from a reference image plus a prompt, or from a Pop-Cam style preset.
1 credit per image, charged only when an image comes back — credits provided by Pop-Cam, the same maker's image app, so there is no Pop-Cam account or key to set up. A reference keeps a character, product or face the same across every shot of a video, and a 9:16 reference gives a 9:16 image. The local server saves the PNG into dir (a local file works as the reference); the hosted one takes a reference link and returns the image in the chat. Its link lasts an hour.
Promotion videos
Automated Content as tools: describe your product and get back a finished promotion video, ready to post — the same queue as the Automated Content page.
A finished 30–40 second vertical promotion video for your product, made from a brief: the script, AI stills, an AI voiceover, music, captions and an end card with the name and website.
10 credits, taken when the video is queued and refunded automatically if it fails. It takes about 20 minutes, longer when other videos are ahead of it in the queue, and you can have two in the queue at once. Give just your website and Hookova reads it, or a name and a description; the video claims only what those say. Pick a look (3D cartoon, photo-real, anime, watercolor or pixel art) and add a picture of the product, a mascot or yourself if you like — a file on the local server, a public link on the hosted one. English only.
Your promotion videos: where each one is in the queue, how far along it is, and the finished video.
A finished video comes with links to the MP4 and its cover, good for two hours, and a suggested caption for the post. On the local server, dir saves each finished MP4 into a folder. The Hookova web app also tells you when one is done. Free.
Publishing
The Publish page’s three steps as tools: which accounts are connected, post a video or photos to them, and see where each one landed. These are the tools that touch the outside world, and the assistant is told so in each description.
The social accounts connected on the Publish page: id, platform, handle.
YouTube, TikTok, Instagram, Facebook, X, LinkedIn, Pinterest, Threads and Bluesky, each with the name of the group it is in. Connecting one is a sign-in on the platform's own consent screen (Bluesky: a handle and app password), so it happens on hookova.com/publish under Connections, not here.
Your groups of accounts — a brand, a client, a language — each with its accounts, plus the ones in no group.
The same groups as the Groups tab on the Publish page. An account is in one group at most. A group's name is what publish_post and list_posts take as group.
Makes a group, and moves the accounts you name into it.
An account already in another group moves out of it. Posts nothing.
Renames a group, or moves accounts in or out of it.
group is its name or id. An account taken out goes to no group. Posts nothing.
Deletes a group.
Its accounts stay connected and go to no group; no post is touched.
Uploads one video or up to ten photos from your machine and posts them with one caption.
Now, at an ISO time with a timezone offset, or as a draft. The assistant is told to list your accounts and ask which ones, not assume all of them; group posts to every connected account in one of your groups. A photo post skips YouTube. The file goes straight to the scheduler's storage, not through Hookova. This really posts, so the assistant is told to confirm the caption, the accounts and the time with you first. The queue's next slot lives in your browser, so the tool takes a time rather than picking one.
Every post made through Hookova: drafts, scheduled, posting, done — and where each landed.
group narrows to posts that go to an account in that group. Per-account results ride along for the most recent done posts: the URL on each platform, or its error.
One post: caption, status, time, accounts — and, once it has gone out, where it landed.
Changes nothing.
Edits a draft or a scheduled post: the caption, the YouTube title, the accounts, the time.
A draft stays a draft — editing never sends it. A scheduled post keeps its time unless given a new one; draft:true takes it back to drafts. The media stays; to change it, cancel and post again.
Sends a draft, now or at a time.
Only a draft, and only one with a file on it. Really posts.
Withdraws a scheduled post before it goes out, or deletes a draft.
A post already sent is recalled on the platform, not here. No undo.
Sound effects and music
The free sound-effect library and the royalty-free music library, so an assistant building a video can name the sting or the bed it wants and get the file.
Search about a thousand short sound effects — booms, whooshes, stings, meme clips — by what they sound like.
No sign-in. list_sounds browses, list_sound_groups names the sixteen kinds, get_sound reads one.
Put the MP3s in a folder on your machine and answer with the paths, ready for an editor or ffmpeg.
Local server only — the hosted one has no disk, so its results carry each MP3's link instead.
1,442 royalty-free tracks: search by mood, use and length, or describe the video and get tracks ranked by fit.
No sign-in, except recommend_music. list_music browses, list_music_tags lists the moods and uses, get_track reads one.
Put the tracks in a folder and answer with the paths.
Local server only, like the sound downloads.
Color LUTs
The looks from the free LUT photo editor, so an assistant can pick a grade for a clip by name and, on the local server, apply it.
The 43 free sRGB color looks from the LUT photo editor, by name or family, each with its .cube link.
No sign-in. Load the .cube in any editor that takes one; these are SDR looks, not log or HDR conversions.
Save one .cube to a path on your machine.
Local server only. An existing file is never replaced unless you say overwrite.
Grade a photo or video on your machine with FFmpeg, at any strength from 0 to 1, into a new PNG/JPG or MP4.
Local server only; needs FFmpeg and ffprobe on PATH. Free, nothing uploaded, the original never touched. Up to 1 GB, 16 megapixels, 180 seconds.
update_savereplaces the tag list wholesale. If you ask for “add Recipes to that one,” a good client reads the save first and sends the old tags back with the new one — but it’s worth saying “keep the existing tags” out loud the first few times.Let the agent install it
If the assistant you’re installing this for can run shell commands — Claude Code, Cursor’s agent, anything with a terminal — you don’t have to read the rest of this page. Hand it one URL and it fetches the commands itself.
install www.hookova.com/install/mcpThat address serves this same install as plain text, written for whatever is holding the shell rather than for a reader: the zip to fetch, where to put it, the claude mcp add line with the path already absolute, the JSON block for every other client, and where to stop and ask you for a token. Read it yourself with curlif you’d rather see what you’re authorising first — it’s about two hundred lines and it does nothing on its own.
curl https://www.hookova.com/install/mcpIt will stop before signing in.The runbook tells the assistant to install and register the server and then ask you for a token, because that is the one step it can’t do on your behalf — and it says, in as many words, not to ask you for your password. See the runbook and step 1 below for the two token shapes.
Set it up in three steps
The manual path, for a client with no shell of its own — or for anyone who’d rather type it themselves.
You need Bun installed — the server is a TypeScript file run directly, with no build step. Everything else is copy-paste.
Get a token
connect hands you a link, you approve it in the browser, and the key arrives on this machine without passing through the conversation. Nothing to edit, nothing to restart. The rest of this step is the manual route, for a headless machine with no browser or a token you already have.For any account, open Connect to My AI and choose Copy prompt. Paste it into your coding assistant. The token is included. For manual setup, use the block’s HOOKOVA_TOKEN value.
Either way the token lasts 30 days. Step 3 shows how to make renewal automatic.
Unzip and install
Unzip it somewhere you intend to keep — your client will run the file from wherever it lands, so the Downloads folder is a bad choice.
unzip ~/Downloads/hookova-mcp.zip -d ~/hookova-mcp
cd ~/hookova-mcp
bun install
bun link
hookova --helpbun install fetches the dependencies. bun link adds thehookova and hokova commands. Run hookova login to sign in or hookova projects to list boards. If the command is not found, add the directory from bun pm bin -g to your PATH. See the CLI guide for setup and examples.
Register it
In Claude Code, one command does it. Use an absolute path — the client may not expand ~.
claude mcp add hookova \
--env HOOKOVA_TOKEN=eyJ… \
-- bun /Users/you/hookova-mcp/index.tsIf your account has a password, add it here instead of re-pasting a token every month — the server re-mints once on a 401, so expiry becomes invisible:
claude mcp add hookova \
--env HOOKOVA_TOKEN=eyJ… \
--env HOOKOVA_EMAIL=you@example.com \
--env HOOKOVA_PASSWORD=… \
-- bun /Users/you/hookova-mcp/index.tsRestart the client, then check it took: /mcpin Claude Code lists the server and its forty-five tools. Or just ask — “am I signed in to Hookova?” runs auth_statusand comes back with the account, the plan and what’s left today. If you skipped step 1, say “sign in to Hookova” instead and paste the token when it asks.
Other MCP clients
Every other client wants the same three facts in JSON: the command, its arguments, and the environment. Put this in claude_desktop_config.json (Claude desktop app) or .cursor/mcp.json (Cursor), and the same block works in most editor integrations.
{
"mcpServers": {
"hookova": {
"command": "bun",
"args": ["/Users/you/hookova-mcp/index.ts"],
"env": { "HOOKOVA_TOKEN": "eyJ…" }
}
}
}If the client is a desktop app, give command an absolute path too — e.g. /opt/homebrew/bin/bun. A GUI application doesn’t inherit your shell’s PATH, which is the single most common reason a server registers fine and then shows no tools.
Environment variables
| Variable | What it does | |
|---|---|---|
| HOOKOVA_TOKEN | Optional | A minted bearer token or your site session cookie — the server accepts either, and both expire after 30 days. Generate one on the MCP server page in Settings, or mint one with your password. Set here it wins over a saved sign-in at startup; leave it out and sign_in supplies the token instead. |
| HOOKOVA_EMAIL + HOOKOVA_PASSWORD | Optional | Re-mints a token once on a 401, so expiry is invisible instead of a monthly re-paste. Does nothing on a Google-linked account, which has no password to mint with. |
| HOOKOVA_API | Optional | Defaults to https://www.hookova.com. Point it at http://localhost:3000 to work against a dev server. |
| HOOKOVA_TOKEN_FILE | Optional | Where sign_in saves its token. Defaults to ~/.hookova/mcp.json, written at mode 0600. |
Troubleshooting
Every tool answers “Sign in to save.”
Ask for auth_status — it says whether a token is there at all, and if one is, why it was refused and which of the three places it came from. The fix is usually sign_in with a fresh token, which takes effect immediately. If you pasted a session cookie that DevTools had split into __Secure-authjs.session-token.0, .1 and so on, the values must be joined in order with nothing between them.
sign_in worked, but the next session is signed out again.
Either it ran with remember: false, or HOOKOVA_TOKEN is set in your client's config — the environment variable wins at startup, so a stale one there overrides the token sign_in saved. sign_in's answer warns about this when it applies. Remove it from the config and the saved sign-in takes over.
The client starts but lists no tools.
The command couldn't run. Check that bun is on the PATH the client uses (a GUI app doesn't inherit your shell's PATH — give the absolute path to bun, e.g. /opt/homebrew/bin/bun) and that the path to index.ts is absolute, not ~-relative, in JSON configs.
Cannot find package '@modelcontextprotocol/sdk'
bun install wasn't run in the unzipped folder, or was run somewhere else. cd into the folder that holds index.ts and run it there.
A save comes back with no title or synopsis.
That's expected — save_link returns as soon as the row exists. The worker fills in the rest a few seconds later; call get_save again to read it.
“Daily save limit reached.”
The allowance is per account and shared with the phone and the browser extension — there's no separate MCP quota. It resets daily.
From a board to a hook
A Hookova project is a board: a name, a brief in your words, and the saves you pinned to it — some of them clips Hookova has taken apart shot by shot. get_hook_guide writes that board out as HOOK-GUIDE.md, plus a folder of reference stills, and a Claude Code skill hands it to HyperFrames, which renders video from HTML, to build the opening seconds of your video.
The skill ships inside the server folder, but Claude Code only reads skills from ~/.claude/skills, so link it once:
ln -sfn "$HOME/.hookova/mcp/skills/hook-from-project" "$HOME/.claude/skills/hook-from-project"Then, in a fresh directory, say /hook-from-project <board name>. It lists your boards, runs hyperframes init, writes the guide and the stills into the project, asks which pattern to build on when a board has several, and hands the build to the hyperframes skills (npx hyperframes skills updateinstalls them). It spends no credit and never starts an analysis: a board with no analysed clip is told so, and the choice — analyse one first, or build a typographic hook from the subject alone — is yours.
What it can’t do
Stated plainly, because a tool list that overpromises wastes more of your time than one that’s short.
- No separate quota. Saves made here draw on the same daily allowance as the phone and the extension. There is no MCP tier to raise.
- The hosted server can’t take files.Posting a video from your computer, and writing a hook guide’s stills to a folder, need the local server.
- No images or video in answers.Keyframes and source videos are auth-gated URLs a model can’t open, so
get_savedrops them rather than handing back links that 401. The one exception is on disk:get_hook_guidewith adirfetches the reference stills itself and writes them torefs/. - Bulk export has a floor.
export_savesreads the Obsidian export route, which returns one page of at most 2,000 rows with no cursor. Past that, the oldest saves aren’t reachable through this tool — uselist_saveswith a search instead. - No starting an analysis. Shot-by-shot breakdowns are run in the web app;
get_savetells you when one exists for a saved URL, andget_hook_guidereads the finished ones on a board. - No revocation.
sign_outforgets a token; it cannot cancel one. There is no revocation endpoint behind it, so a token that has leaked stays usable until its 30 days run out.
Frequently asked questions
Can my assistant install this for me?
Yes, if it can run shell commands. Say "install www.hookova.com/install/mcp" — that URL serves the whole install as plain text written for an agent rather than a reader: the zip to fetch, where to put it, the claude mcp add line with the path already absolute, and the JSON block for other clients. It will install and register the server, then stop and ask you for a token, which is the only step it can't do for you. Read it first with curl if you'd rather see what you're authorising.
Does saving through MCP spend a credit?
No. Saving is free. Credits pay for analyses, transcripts, AI images and AI videos; saves count against your daily save allowance, which is shared with the iOS app and the Chrome extension — there is no separate MCP quota.
Do I need the Hookova source repository?
No. The zip is the whole server: one TypeScript file, a package.json, a tsconfig and a README. It calls the public API over HTTPS like any other client.
Where does my data go?
Nowhere new. The local server is a process on your machine that holds one token and talks to the same Hookova API the iOS app has used since it shipped; the hosted one runs on hookova.com itself and calls those same routes. Either way your assistant sees what the tools return, and nothing else — no tool ever returns a token.
Is it safe to give my password to sign_in?
Prefer not to. A password passed as a tool argument goes into the conversation transcript wherever your client stores it, the same as anything else you type. Use the token argument instead, or put HOOKOVA_EMAIL and HOOKOVA_PASSWORD in the client's config — the server re-mints from those on a 401 without either value ever entering the conversation. The password path exists because it is sometimes the quickest way in, not because it is the best one.
Does sign_out revoke the token?
No. It forgets the local server's token — clears it from the running server and deletes the saved file — but a 30-day token can't be revoked, so it stays valid until it expires. Treat one that has leaked as still live. Hosted connections are different: Disconnect under Settings → MCP server ends one for good within the hour.
My account signs in with Google. Can I still use it?
Yes. The hosted URL is the easy way: your app sends you to Hookova in your own browser, where Google sign-in works as usual. For the local server, Google-linked accounts have no password to mint a token with — use the connect tool, or generate a token on the MCP server page in Settings and paste it into your config. That one expires after 30 days, and generating a fresh one is the renewal.
Which clients does it work with?
The hosted URL works with anything that can add a remote MCP server with OAuth sign-in: the Claude app and claude.ai, Claude Code, ChatGPT, Cursor and most current clients. The local server works with anything that speaks MCP over stdio. Both reach the same account.
Can it analyze a video, not just save it?
It can't start one — shot-by-shot analysis runs in the web app and spends a credit there. It can read the finished ones: get_save tells you whether an analysis exists for a URL you've saved, and get_hook_guide turns every analysed clip on a project board into a hook pattern, written out shot by shot.
Can it help me make the video, not just study other people's?
That is what the hook guide is for. Pin the clips you want to learn from to a project, give it a brief, and get_hook_guide writes the board out as HOOK-GUIDE.md with reference stills. The hook-from-project skill hands that to HyperFrames, which builds the opening seconds of your own video on the pattern of those clips — their timing, framing and technique — with your brief as the subject. Nothing from the source clips is reused, and no model runs inside Hookova to make the guide: it is your own analyses, re-shaped.
Grab the zip, paste the three commands, and the next thing you read worth keeping can be saved without leaving the conversation you’re in.