Set up Collections with AI

Describe your client's content in plain language and let Cursor, Claude Code, or another MCP client design the collections, import the content, and wire up the site.

The quickest way to set up collections is to describe the content to your AI coding assistant. With the Dropl MCP server connected, it designs the fields, shows you the plan, creates the collections after you approve, imports the content your site already has, and replaces the hard-coded content with live data from Dropl.

You stay in control: nothing is created until you approve the plan, and changes that could lose data need your explicit go-ahead.

1. Connect Dropl to your assistant

Follow Set up with AI to add the MCP server and sign in. Leave the three Collections permissions ticked when you approve the sign-in:

Permission	What it allows
`collections:read`	Read collections, their fields, and items (drafts included)
`collections:write`	Add, edit, reorder, and delete items
`collections:schema`	Create collections and change their fields

A key without collections:schema can manage content but can never change a collection's structure.

2. Write a brief

Describe what your client needs to edit, the way you'd explain it to a developer:

What the items are: dishes, machines, courses, projects, jobs.
What each item has: name, description, price, photos, dates, links.
Fixed choices: categories or tags, with their options (for example starters, mains, desserts).
Yes/no switches: sold out, featured, available.
What's required, and anything your client might get wrong, so the assistant can add help text.

You can also leave the brief short and point the assistant at the page where the content is hard-coded. It reads the code and suggests the fields.

In the dashboard, open Collections and click Set up with AI, then pick Make content editable (Collections) to fill in a brief and copy a complete prompt.

3. What the assistant does

Reads first. It lists the site's collections and reads the current fields of any that already exist, including edits you or your client made in the dashboard.
Plans. It designs the collections (fields, types, options, required fields, the title field, and friendly help text) and runs plan_collections. This is a dry run: nothing is saved.
Asks you. It shows you the plan as a plain-text summary, like Add field "Spice level" or Remove field "Notes": 12 items lose their value.
Applies. After you approve, it creates or updates the collections with apply_collection_plan. Up to 10 collections can be planned and applied at once.
Imports. It finds the content already in your project (arrays, JSON files, repeated markup) and imports it with add_collection_items. It dry runs first and shows you any items that don't fit the fields.
Wires up the site. It calls get_collection_code for TypeScript types and a fetch example, and replaces the hard-coded content with live data, following your project's conventions.
Checks its work. It runs your build and fixes what it broke.
Hands off. Invite your client from Settings → Sites so they can edit the content from their phone.

Changes need your approval

Adding fields is always safe. Renaming or removing a field or option, or changing a field's type, can lose data or confuse the people editing it, so it's destructive. Dropl refuses destructive changes unless the assistant passes confirmDestructive, and the assistant is told to do that only after you explicitly agree.

If you or your client changed the collection in the dashboard after the assistant read it, Dropl refuses the plan as a conflict instead of overwriting your edits. The assistant reads the fields again and builds a new plan. Your manual edits always win.

Applying the same plan twice doesn't create duplicates, and retrying an import doesn't add the same items twice.

Made a change you didn't want? The assistant can undo the most recent schema change with undo_collection_change, or you can click Undo on the collection's Activity tab, as long as no data would be lost.

Example briefs

Copy a prompt below and paste it into your assistant's chat, then adjust the details for your client.

Use the Dropl MCP tools to make this site's content editable with Dropl Collections.
 
This is a restaurant site. The client needs to edit the menu: dishes with a name, description, price, a photo, and a category (starters, mains, desserts, drinks), plus dietary tags (vegan, gluten-free, spicy) and a sold-out toggle. They also post weekly specials with a start and end date.
 
Read the client site's existing collections first. Show me the plan from plan_collections and wait for my confirmation before applying it. Then import the content that's hard-coded in the project (dry run first), replace it with live data using get_collection_code, and run the build.

Equipment inventory

Use the Dropl MCP tools to make this site's content editable with Dropl Collections.
 
This is an equipment rental and sales site. The client needs to manage their inventory: each machine has a name, make, model, year, hours of use, condition (new, used, refurbished), price, a short description, several photos, and an available toggle. Visitors filter by category (excavators, loaders, trailers, attachments).
 
Read the client site's existing collections first. Show me the plan from plan_collections and wait for my confirmation before applying it. Then import the content that's hard-coded in the project (dry run first), replace it with live data using get_collection_code, and run the build.

Training course schedule

Use the Dropl MCP tools to make this site's content editable with Dropl Collections.
 
This is a training provider's site. The client needs to publish upcoming courses: course name, description, start date and time, end date and time, location (or online), instructor, price, seats available, a booking link, and a level (beginner, intermediate, advanced). Past courses should be easy to hide.
 
Read the client site's existing collections first. Show me the plan from plan_collections and wait for my confirmation before applying it. Then import the content that's hard-coded in the project (dry run first), replace it with live data using get_collection_code, and run the build.

Contractor project portfolio

Use the Dropl MCP tools to make this site's content editable with Dropl Collections.
 
This is a home renovation contractor's site. The client needs to add finished projects: project name, town, a description, completion date, project type (kitchen, bathroom, deck, addition), a cover photo and a few more photos, the budget range, and a featured toggle for the home page.
 
Read the client site's existing collections first. Show me the plan from plan_collections and wait for my confirmation before applying it. Then import the content that's hard-coded in the project (dry run first), replace it with live data using get_collection_code, and run the build.

Job listings

Use the Dropl MCP tools to make this site's content editable with Dropl Collections.
 
This is a company careers page. The client's HR team needs to post open roles: job title, department, location, employment type (full-time, part-time, contract), salary range, a description with responsibilities and requirements, an apply link, a closing date, and a remote-friendly toggle.
 
Read the client site's existing collections first. Show me the plan from plan_collections and wait for my confirmation before applying it. Then import the content that's hard-coded in the project (dry run first), replace it with live data using get_collection_code, and run the build.

The Collections tools

Tool	What it's for
`list_collections`	A site's collections, item counts, and plan limits
`get_collection_schema`	A collection's current fields and version
`plan_collections`	Preview creating or changing collections, without saving
`apply_collection_plan`	Apply an approved plan
`add_collection_items`	Import items, with errors for each item that doesn't fit
`list_collection_items`	Search and filter items
`get_collection_code`	TypeScript types and a Next.js example for the website
`undo_collection_change`	Undo the most recent schema change

Troubleshooting

"Someone changed this collection since you read it." Fields were edited after the assistant read them. Ask it to read the collection again and make a new plan.

The assistant is missing a permission (INSUFFICIENT_SCOPE). The connection was approved without a Collections permission. Run npx -y @dropl/mcp login again and leave the Collections permissions ticked.

Some items weren't imported. The import reports each item's problem by field. Ask the assistant to fix those items and import only them again.

The plan's item limit is reached. Existing items keep showing on the website. Remove items you don't need, or add more from Billing.

Next steps

Show collections on your website: what the assistant's code calls, if you want to write it yourself.
Collections: fields, items, imports, and what clients see.