How to Use Boost.space to Clean Product Data for AI Agents

The short answer: create a space for products with the SKU as its key field, import your catalog file and check the preview, connect your shop through a Make scenario in the Connect view, add an AI field that drafts short descriptions on demand, review them, then connect an agent over MCP with a token scoped to that data and push approved records back with a Share scenario. If your listings, marketplace feeds and AI-search visibility all suffer from the same messy catalog, this is the order Boost.space's docs set out, and the entry Agent OS plan holds 100,000 managed records for $300 a month paid yearly, checked on 24 October 2026 on the pricing page.

Disclosure: I may earn a commission through some links on this site. It does not change what this page says.

An agent is only as good as the records it reads. Boost.space's whole pitch is that the data comes first, and for a catalog that is right: an agent writing product copy from three conflicting price fields will pick one, confidently. For what the product is and what it costs, see the Boost.space glossary entry.

What to prepare:

1. The record count, worked out. Agent OS plans are sized in managed records (pricing page). For example, 12,000 SKUs arriving from a shop, a supplier feed and a spreadsheet = 12,000 records if the key field merges them, and 3 x 12,000 = 36,000 if it does not.
2. A catalog export you already have. A CSV, XLSX or XML file from your shop or PIM, with one stable identifier per product, usually the SKU.
3. The before number. Count duplicates and empty description fields in that export now, so you can compare after the first import and again after a month.

1. Create the space and set its key field

Sign in to the app, create a space in a module for products, and choose the key field in the space wizard (quickstart, key fields). When data arrives from an import, a scenario or the API, Boost.space compares the key field: a match updates the existing record, no match creates a new one. Pick the SKU, not the product name, which changes when someone fixes a typo.

2. Import the catalog file

Open the module, choose the space and click + New Data source, then Upload a file (import guide). Boost.space analyses the file, proposes field types with AI, and lets you map each column to an existing field or a new one. Read the data preview before you continue: it shows the mapping and flags invalid values, and only valid data is imported.

3. Connect the shop through Make

Automation between Boost.space and other apps runs on Make, on your own account. In Make, create an API token; in Boost.space, paste it under System settings, Integration and choose your team or organization (views). The module then shows views for admins: Connect for scenarios that import data, Enrich for ones that add to records, Share for ones that export, and Flow for the whole picture. Add a Connect scenario that pulls new and changed products from the shop.

4. Decide which source wins

When the shop and the supplier feed disagree on a field, consolidation decides: the key field matches the records, and API token prioritization sets which source's value wins (key fields). Both are set from the Flow view. Make the shop win on price and stock, the supplier on dimensions and materials, and write the rule down for the team.

5. Draft descriptions with an AI field, under review

In System settings, open the products module, click New field and choose the AI type (AI fields). Write a custom prompt that inserts other fields, for example "Write a two-sentence description of {Name} for {Category}, using only {Materials} and {Dimensions}." Set the execution mode to Manual, so values are generated on demand rather than regenerated every time a source field changes, and choose tone, length and formatting in the advanced settings.

This is where the human check sits. The AI field drafts; a person on the e-commerce team reads each draft against the product before it is marked ready. AI credits come with the plan each month, are used only by AI features, and do not roll over (AI credits).

6. Connect an agent over MCP, with limited rights

Open the Connect AI window from the custom module's Connect AI button, choose MCP token, pick your client and copy the connection URL it gives you rather than building one by hand (Connect via MCP). An agent connected this way acts with exactly the creating user's permissions, so create the token from a user who can see the products space and nothing else (tokens). The agent can search, read, create and update records one at a time; it cannot delete records or write in bulk over MCP. Ask it for proposals, such as products with missing materials, and have a person accept the changes.

7. Share approved records back

Add a Share scenario that exports records marked ready to the shop or a marketplace feed (views). Share exports data out, so it has no consolidation settings; Connect and Enrich are the scenarios that write into your records, which is why step 4 matters.

8. Measure in Billing

Administrators see records, AI credits, operations and admin seats in the Billing section, from the homepage or Profile settings, Billing (plans guide). Compare duplicates and empty descriptions with the before numbers from your export.

What the docs warn about

• Tokens are passwords. Anyone holding an MCP or API token can act on your data up to its permissions (tokens).
• A missing key field duplicates everything. Without one, every import and sync creates new records instead of updating (key fields).
• Views only appear for admins on a Make-connected module. Everyone else sees the data view (views).
• Inline AI cannot use your own key. Bring your own key works for automations, but inline database features use AI credits (AI credits).

Agent OS is priced on managed records, so the vendor gains as the catalog grows inside it. The plans guide's own advice is to delete old records to free room (plans guide); archive discontinued products elsewhere before they count.

FAQ

Do I need a Make account?

For moving data in and out of other apps, yes: scenarios run on your connected Make account. Reacting to record changes or schedules inside Boost.space works with native triggers and actions or Python code automations (automation).

Can I import from Excel?

Yes. The import accepts XLSX as well as CSV and XML (import guide).

Which AI assistants can connect?

Claude, ChatGPT, Cursor, VS Code and any client with a custom MCP connector field, either with the URL and an MCP token or through OAuth (Connect via MCP).