acqai

ACQ AI, in Claude.

You are in a Claude Code or Cowork conversation about your business. This plugin lets that conversation include ACQ AI. Claude reads the docs you point it at, writes the question, shows it to you, sends it to ACQ AI through your own signed-in browser once you say yes, reads the answer, asks a follow-up or two with what your docs add, and hands you the result with the edits it implies. Your docs are the context, so there is nothing to paste and no context document to keep current.

This is its own GitHub marketplace (scottsolowild/acqai-plugin), not Anthropic's plugin catalog. You will not find it by searching Claude's official list.

Install (pick one)

Option A: Terminal

claude plugin marketplace add scottsolowild/acqai-plugin
claude plugin install acqai@acqai-plugin

If you belong to more than one company on ACQ AI:

claude plugin install acqai@acqai-plugin --config mozi_company="Your Company"

Option B: Claude Code Desktop (local session)

Cloud sessions do not load this plugin. Use a local Code session on your machine.

1. Open Claude Code Desktop → a local project (a folder of docs is enough).
2. + → Plugins → Add marketplace.
3. URL: scottsolowild/acqai-plugin → Sync.
4. Install acqai from that marketplace. Confirm it shows enabled under + → Plugins.

If Sync shows only Failed to add marketplace, use Option A instead. Do not type /plugin … in the Desktop composer; that command is not available there. If you already ran Option A, skip Add marketplace. The plugin is already under + → Plugins. Quit Desktop and reopen if it is missing, so it reloads ~/.claude/settings.json.

Option C: Cowork (Claude desktop app)

A Cowork task that runs on your computer loads it. A Cowork task that runs in Anthropic's cloud does not.

1. In the Claude desktop app, open Customize → Plugins.
2. Add marketplace → scottsolowild/acqai-plugin.
3. Install acqai from it.

In Cowork the command is /acqai:acq. If it says the plugin's connector did not start, open acqai under Customize → Plugins and connect it on its Connectors tab. A plugin installed here is saved to your Claude account, so Claude Code picks it up at its next session start too.

After install (once, a few minutes)

1. Type /acq in a Claude Code chat (Desktop local, or claude in a terminal), or /acqai:acq in Cowork, or say "set up ACQ AI". On its own, /acq gets you ready: Claude installs the browser it drives (a private copy in ~/.config/acqai) and opens a sign-in window at portal.acquisition.com/advisor.
2. Sign in in that window. Email code, as usual. If a company list shows, click yours. Send one short message there ("hi" is enough). The plugin learns the chat route from that message and closes the window.
3. Try it. Say ask ACQ AI what it would change first about my offer and point it at a doc.

From a terminal, the same chat looks like:

claude

Then type the ask. Or one shot:

claude -p "ask ACQ AI what it would change first about my offer"

Use it

Say what you want in plain English:

• /acq on its own: get ready. It checks you are signed in, opens the sign-in window when you are not, and says what is next. Every ask runs the same check first — a question after /acq does not skip it.
• "ask ACQ AI to price this offer" (with the offer doc open or named)
• "run my landing page past ACQ AI and tell me what it would change"
• "what would ACQ AI say about this funnel? here are the numbers"
• "ask a follow-up: how does that change at my price?"
• "sort that reply: what do we adopt, what waits, what do we drop"
• "go back to yesterday's pricing conversation and ask about the pilot"
• "keep Jane Doe out of anything you send"

Or from the slash command:

• /acq <task> — show each question, wait for yes in chat, send, ask again before each follow-up, then wait before editing files
• /acq -y <task> (or --yes) — the flag is the yes for the whole run: show the first question, send it and up to two follow-ups, write the Adopt edits (new files when the task calls for them), and record the outcome, with no further ask

Claude writes the question from your docs and shows it before anything is sent. With -y, that show is for the record; the flag already covered the send and the file edits. Each conversation is kept in ~/.config/acqai/answers/ as one file, so nothing is lost when the chat scrolls away. The file opens on the question, the best answer, and what changed, and each message below it sits under who sent it.

What you need

• A paid ACQ AI (Mozi) account you can sign into in a browser.
• Claude Code (the terminal, or the desktop app's Code tab on a local session), or Cowork in the Claude desktop app, on a task that runs on your computer.
• Python 3.9 or newer. A Mac already has it.

Cursor is not required. A repo is not required. A folder of docs is enough, and no folder at all still works: Claude sends your question on its own.

What it does

• Writes the question from your docs. The situation, the numbers, the decision, and the docs that matter, pasted in whole, with a standing ask that ACQ AI use your terms and say what each recommendation rests on.
• Sends it through your own browser. Playwright drives a private Chromium that you signed into once. The session stays on your machine.
• Refines with questions. Claude reads the answer and asks a follow-up or two in the same ACQ AI conversation, each carrying a number or a result from your docs that ACQ AI did not have yet. Each answer builds on the last, so "go deeper on point two" works too. A new topic starts a new conversation.
• Brings the mechanics home. Claude labels what ACQ AI said and what it added, then sorts the answer: adopt (an edit, drafted in your voice), later, or drop. What you decide is recorded on the answer's file, so the list of answers shows which ones you acted on.

The rules

• Your yes before anything leaves. The script asks before it sends, and Claude asks you before it runs the script. /acq -y is one yes for the first question and its follow-ups in that run, not a silent batch.
• One question at a time on the wire. The script paces itself.
• Your account, your use. It does what you would do by hand, in your own browser, for yourself.
• Chat only. It never posts to the community.
• Private things stay home. Client names, call transcripts, anything you call private. Claude leaves them out, and you see the question before it goes. Names you add to the private list are checked on every send. When a question includes one, the send stops before it goes out.

Questions

Is Cursor required? No. It runs from Claude Code in a terminal, from the desktop app's Code tab on a local session, or from Cowork on your computer. Cloud sessions do not load this plugin. Cursor is one editor that can host Claude Code, and it is not needed.

Why does it come with a connector? Cowork runs Claude's commands in a Linux machine of its own on your computer, with no screen to show you the sign-in window and no route to ACQ AI. The connector runs the same commands on your computer itself, where your browser and your login are. Claude Code uses the connector too, so both apps run one path.

Do I need to know what a repo is? No. Point it at a folder of docs, or at one doc, or at nothing.

Is there an ACQ AI API? No, and none has been announced. So the plugin uses your own signed-in browser, on your own computer, the way you would by hand. That is also why it runs on your machine and not in a cloud session.

Is this an official ACQ tool? No. One member built it and shared it. It uses your account through your browser. If ACQ asks you to stop, stop.

What does it read? The docs you point it at, and the question you approved. Nothing else leaves your machine, and nothing goes anywhere but ACQ AI.

Where is my login kept? In ~/.config/acqai/browser-profile, on your computer. Delete that folder to sign out.

Will it work on Windows or Linux? It was built and is used on a Mac. Playwright runs on all three, so the pieces are there, and nobody has run it on the other two yet. If you do, say how it went.

The sign-in window showed a list of companies. Click yours. To have it clicked for you next time: acqai.py setup --company "Your Company".

It says the route is not learned. Run login again and send one message in the window. If that still does not take: in Chrome, open portal.acquisition.com/advisor, open DevTools → Network, send any message, right-click the chat-stream request (Create is …/sandbar/chat; the stream is what carries the answer), Copy → Copy as cURL, save it to a file, and run acqai.py discover --from-curl that-file. The cookie in it is read and thrown away. Only the route is kept.

I still use the older ai.acquisition.com app. Run acqai.py login-legacy. Send one message in that window so the learned route matches. Later sends follow that route's host. login alone returns you to portal. A conversation continues only on the app that started it, so a send on the other app stops before it goes out. It says that --new starts a fresh conversation there, and which login goes back.

Run it by hand (optional)

The skill runs the bundled script for you, through the plugin's connector. To run it yourself, from a checkout of this repo:

python3 plugins/acqai/scripts/acqai.py setup --company "Your Company"
python3 plugins/acqai/scripts/acqai.py login
python3 plugins/acqai/scripts/acqai.py probe
python3 plugins/acqai/scripts/acqai.py ready
python3 plugins/acqai/scripts/acqai.py send "What would you change first about this offer?"
python3 plugins/acqai/scripts/acqai.py send --file question.md -y
python3 plugins/acqai/scripts/acqai.py send "Go deeper on point two."
python3 plugins/acqai/scripts/acqai.py send "Context first." --paste
python3 plugins/acqai/scripts/acqai.py answers
python3 plugins/acqai/scripts/acqai.py show 2026-09-23-2209
python3 plugins/acqai/scripts/acqai.py send "And the pilot?" --continue 2026-09-23-2209 --why "The pilot is priced already."
python3 plugins/acqai/scripts/acqai.py names add "Jane Doe"
python3 plugins/acqai/scripts/acqai.py outcome 2026-09-23-2209 --answer "- Lead with the pilot." --adopt "Lead with the pilot."
python3 plugins/acqai/scripts/acqai.py answers --regroup --dry-run

ready checks what a send needs and fixes what it can: setup when Playwright is missing, the sign-in window when the login is out, then the route and the chat the next send continues. --dry-run names the steps and runs nothing. send asks y/N before it sends. -y answers it. --dry-run shows what would go out and sends nothing, and it exits 1 when a line says NOT ready. --new starts a fresh conversation, and --continue goes back to the one an answer on file used. --why keeps a follow-up's reason above it in that file. --paste takes the question from the clipboard, after an optional note. show prints one answer's whole file. names lists the private names. outcome reads or records the top of an answer's file, the best answer and what came of it (adopt, suggest, later, or drop), and answers --regroup merges the one-file-per-send answers of earlier versions into one file per chat.

Config, all optional:

Variable	Default	Meaning
ACQAI_STATE_DIR	~/.config/acqai	The folder that holds the login, the route, the chat id, the private names, the venv, and the answers.
MOZI_COMPANY	(saved by setup --company)	The company to click on the sign-in list.
MOZI_BASE	(learned route, else portal)	Override the app origin. Prefer login / login-legacy. Re-run login after switching.
MOZI_MIN_DELAY, MOZI_MAX_DELAY	12, 30	Seconds between sends, a random gap in that range.
MOZI_SEND_OK	(unset)	1 gives the yes for the whole shell. -y gives it for one send, which is the better habit.
MOZI_TOKEN	(unset)	The --http path only: the whole Cookie header from a signed-in request. Prefer the browser path; a captured legacy Clerk cookie dies in about a minute.
ACQAI_LOGIN_WAIT	600	How long login waits for you, in seconds.

Changelog

Version history and the versioning contract are in CHANGELOG.md. Released versions are git tags acqai--vX.Y.Z (Claude) and vX.Y.Z. CI stamps each one after a merge to main, from the commits' Conventional Commit types, and publishes it as a GitHub Release. ./release.sh shows what the next release would be.

License

MIT. Take it, change it, ship your own.