Let your agent work your loop.
CatalogLift speaks MCP. Connect the assistant you already use — Claude, ChatGPT, Claude Code, or code of your own — and it works the same loop you do: reads the day, orders drafts against your playbook, decides on them, writes them to WooCommerce. With your authority, through the same gates, leaving the same receipts.
- Your account's reach — never more
- Every draft and decision stamped with the door it came through
- Disconnect in one click, on any plan
You say it. The loop does it. The panel keeps the receipt.
Three of the things people actually ask for. Every tool named below is one this server registers — nothing here is a roadmap.
“What needs attention today?”
The digest from Today, in your own chat. Nothing was written.
get-loop-digest · list-stores · get-store
“Check the Northbay catalog again.”
Watch and Findings refresh exactly as they do after Check now and Scan.
request-catalog-check · get-watch-ledger · start-scan · list-findings
“Fix the thin descriptions the way the playbook says.”
Every draft in Review and every receipt in Proof says it was ordered and decided by your agent. Drafts needing your own eyes stay in the queue.
get-playbook · request-fix-draft-batch · get-review-queue · get-proposal · approve-proposal-batch
The reach is the point.
An agent is worth connecting because the loop already works in batches: one order drafts 50 fixes, one decision settles 200 of them. What changes is who walks it — and whether you were at the keyboard when it happened.
What it can do
Everything you can: read the day, run checks and scans, order drafts against your playbook, approve and reject, dictate a fix, retry an apply, resume a paused queue, roll a change back, and publish a new playbook.
What it cannot do
Go past your plan, reach a store outside your workspace, skip a lock or a stale check, ship without the read-back and re-scan, or spend outside your monthly allowance.
What you see afterwards
Every draft it ordered says so in Review; every change it decided says so on its receipt in Proof and in the product's own file; the Agents screen counts its work day by day.
How you cut it off
One click on the Agents screen, on any plan. Tokens die immediately, including any issued while you were clicking.
Rollback for 150 days
An eligible change an agent made rolls back exactly like one you made, from its own receipt, with the same current-value check first.
One allowance, not two
Drafts an agent orders come out of your workspace's monthly pool. No second bill, no metered usage, and it can read what is left.
Fair questions, straight answers.
Which assistants work?
Anything that speaks MCP: Claude at claude.ai, ChatGPT in developer mode, Claude Code in your terminal, or your own code against the Claude API or OpenAI Responses. You point the client at one address, sign in with your CatalogLift account, and it discovers the rest.
Does the agent get its own AI budget?
No. Drafts it orders come out of the same monthly allowance the panel spends — 1,000 AI runs a month on Store, per store you pay for — and it can read what is left at any time. There is no second bill and no metered usage.
What can it do that I cannot?
Nothing. That is the point: an agent has exactly your reach in the loop and no more. It cannot go past your plan, cannot touch a store outside your workspace, cannot skip the read-back or ship work the store has moved past.
Does a person have to confirm what the agent decides?
Not as a gate, and we would rather say so than imply one. Risky writes are priced first: the agent asks what a decision would touch, gets the number back, and confirms exactly that. It protects a yes from landing on a different set of products; it is a consistency check on the agent's own step, not a second pair of hands.
How fast can it write?
Fast enough to clear a morning, not fast enough to run away: 600 writing calls an hour per workspace, shared between every agent you connect — and one call can settle up to 200 drafts, so the ceiling is on runaway scripts, not on a working day. Reads do not count against it, though every request shares the server's own per-minute rate limit. If a script ever meets the ceiling it gets a sentence saying when to try again, not an error.
Can I cut it off?
One click on the Agents screen, on any plan — including after a downgrade. The tokens it holds are revoked there and then, and a token that was already being issued when you clicked dies on arrival rather than slipping through. Disconnecting stops what comes next; eligible changes it already made can be rolled back from their receipts for 150 days after verification, subject to the same value and connector limits as changes made in the panel.
Does it work on the free plan?
No. Free scans your catalog, shows every finding and lets you judge sample drafts in the panel. Connecting an agent is part of a paid plan, alongside writing to WooCommerce at all.
Connect a store first. The agent comes after.
The scan is free and runs on your real catalog. When you are ready to hand the work to an agent, the Agents screen has the address and the steps. Every change keeps a record of who decided it, and when.
No card required for the scan. Connecting an agent is part of a paid plan.