Run Your Directory From Your AI Chat: What the 39 Tools Do
A tour of the MCP tool surface behind a conversational directory, grouped by what you would actually ask for, plus the operations worth doing monthly.
Your AI chat can build this directory.
Describe the niche, watch the agent design the fields and fill the catalogue. Free plan, no card.
You never call these by name. They exist so that when you type "feature the Bordeaux listing until the end of September", your AI client knows what to do.
But it is useful to know what the surface covers, because it tells you what you can ask for. Here is the map, grouped by intent rather than by API.
Structure: setting the thing up
Creating the directory and minting its subdomain. One instruction.
Designing the listing schema. This is the important one and it happens before any content exists. You describe the niche, the agent declares the typed attributes that characterise an entry in that market. For gunsmiths that produces a regulatory weapon category and a partner-armoury delivery flag, fields no generic template ships.
This is the step to slow down on. Everything downstream depends on it and it is close to impossible to retrofit at three hundred entries. Expect to correct the first proposal once.
Categories. Create, rename, re-describe, move under a parent, or reposition several at once.
Collections. Indexable landing pages grouping every published listing that carries a shared trait. "Gunsmiths offering partner-armoury delivery" is a collection, and it is exactly the comparative query that still earns clicks after AI Overviews.
Content: filling it
Adding listings one at a time, or in bulk. Bulk operations match on slug, so re-running an import updates rather than duplicates. That single behaviour is what makes iterative importing safe.
Researching a niche. A background job that discovers entries, extracts their data against your typed schema, and inserts them. Useful for a first pass. Treat the output as a draft, and verify anything a visitor would act on: contact details, certifications, regulatory categories, hours.
Updating and deleting. Individual corrections, which is the operation you will use most over the life of the directory.
Blog posts. Create and update. Your content roadmap should come from your failed searches rather than from guessing, which is covered below.
Operations: running it day to day
Publishing. Take the directory live, or back to draft.
Featuring a listing until a specific date, or clearing the feature. This is the mechanism behind paid prominence, though note that end-user monetisation is currently paused on the platform.
Submissions. Open or close the public submit page, read the queue, approve a submission into a draft listing, or reject it. Batching this twice a week keeps it from becoming the whole job.
Themes. Set the public site's design. Deliberately limited. If pixel-level design control is the point, this is the wrong category of tool.
Distribution: getting indexed
Attaching a custom domain, verifying its DNS, listing what is attached, and removing one.
This matters more than it sounds. Free subdomains are noindex by design, so a shared domain does not become an SEO spam magnet that damages every tenant on it. Indexation arrives with the custom domain on the paid plan.
Practical consequence: attach the domain before you build, so your content is indexed on its final URL from the start rather than needing a migration later.
Measurement: the tools worth a monthly habit
Three things to ask for, once a month, in about ten minutes.
Analytics. Searches, clicks, click-through rate over a window. The number that matters is not sessions, it is whether people who search are finding something.
Top listings. Most-clicked entries. Tells you which parts of your set are pulling weight, and which categories deserve more depth.
Failed searches. Queries your visitors ran that returned nothing. This is the single most valuable output on the list.
Failed searches are first-party demand data that no competitor has. Each cluster tells you one of four things: you are missing entries, you are missing a synonym, you are missing an attribute your schema does not model, or there is content demand you have not served. All four are actionable and the second one takes minutes.
The one trap: a failed search does not justify a new page unless you can put five to ten real entries behind it. Otherwise you are generating the thin pages that Google's March 2026 enforcement targeted.
Housekeeping
Jobs. Long-running operations like niche research run in the background. You can list them and check progress rather than waiting.
Subscription. Which plan you are on and what it allows.
What this actually changes
Setup is a few hours of your life, once. It is not where the value is.
The value is that eight months from now, when a listing has a wrong price and you are busy and slightly bored of the project, fixing it is a sentence typed where you already are. The change lands at the next read, around 0.1 seconds, with no rebuild.
Compare that to opening a dashboard, navigating, editing, saving and waiting. One of those happens. The other gets postponed, and postponed corrections are how directory data goes stale, and stale data is the leading cause of directory decline, ahead of competition and ahead of algorithm updates.
That is the entire argument for this architecture. Not speed on day one. Friction on day four hundred.
Where it falls short
Bulk visual review. Scanning three hundred listings for something that looks wrong is genuinely better in a table. Conversation is good at targeted operations and weaker at browsing.
Teams. Each person connects their own client. There is no shared visual admin, which is a real limitation for anything beyond a solo operator.
Design. You choose from what the platform offers.
Monetisation. Currently paused, so the featuring tools exist but the paid flows behind them do not.
FAQ
Do I need to memorise any of this?
No. Describe what you want. The map above is context, not a manual.
Can I do everything without a dashboard?
Yes, and there is no dashboard, so it is not optional.
What if the agent picks the wrong operation?
Correct it in the next message. Changes are visible immediately, which makes mistakes cheap.
Is bulk import safe to re-run?
Yes. It matches on slug and updates rather than duplicating.
Which of these should I use most?
Updating listings, and reading failed searches. The first keeps the directory accurate, the second tells you what to build next.
Connect once and describe your niche →