All jobs

Flag Products in Each Online Store's Airtable Catalog That Are Not Ready to List

Software for online stores makes one agent per merchant that reads their Airtable product catalog each week and lists every product missing a price, SKU or photo.

Run this on RushOpens the console with this prompt ready to send. Connect Airtable there first.

What the agent does

This is for a company that sells software to small online stores: a tool that publishes a merchant's products to their storefront and to marketplaces. Many of those merchants keep their catalog in an Airtable base they built themselves, one row per product, and a product that goes out with no price, a duplicated SKU or no photo is rejected by the marketplace or, worse, goes live wrong. The company saves one agent as a template whose connections name airtable, Airtable's remote MCP server at https://mcp.airtable.com/mcp, and makes one agent from it for each merchant, storing the agent id against that merchant's account.

Each merchant connects their own Airtable once. The company's backend calls POST /o/{org}/agents/{agent}/connections {"type": "mcp", "name": "airtable"} and opens the authorize_url in a popup. On Airtable's own page the merchant chooses which bases and workspaces to share, so the agent never sees a base they did not pick. The software company registers no Airtable app and stores no token.

On a run, the agent finds the catalog with Airtable's search_bases tool, using the base name the merchant gave in the company's settings, reads the field names and types with get_table_schema, and pages through the products with list_records_for_table. It maps the merchant's own field names to the checks in its input: a price above zero, a SKU that no other row uses, at least one image in the attachment field, a description of at least the given length, a weight for shipping, and a stock count that is a whole number and not negative. A row marked draft or archived in the merchant's status field is skipped.

For every product that fails a check, it records the record id, the product name, the SKU and each problem in plain words ("no price", "SKU also used by Linen Apron, Sage"). When the merchant's base has no field for something the checks need, it says so once instead of flagging every row. It changes no record: the template's instructions and the prompt both forbid update_records_for_table and create_records_for_table.

Start it with one API call

POST https://api.phoenix.so/o/{org}/sessions with this body:

{
  "agent": "<agent id>",
  "input": "Find the Airtable base named 'Shop catalog' and its table 'Products'. Read the table schema, then every record. Skip records whose status field is Draft or Archived. Flag a product when: price is missing or 0; its SKU is empty or used by another record; its image attachment field is empty; its description has fewer than 40 words; weight is empty; stock is not a whole number or is below 0. If the table has no field for one of these checks, report that once in missing_fields and skip that check. Return only JSON: {products: [{record_id, name, sku, problems: [string]}], missing_fields: [string], checked: number}. Do not create, edit or delete any record, field or table."
}
  1. Get an API key. In the console, open Settings, then API keys, and create a key for your org. Send it in the Authorization header as a Bearer token.
  2. Send the request. POST the body in request.json to https://api.phoenix.so/o/{org}/sessions, with your repository as the resources URL. The run starts in its own cloud sandbox.
  3. Read what came back. Follow the run on its live stream or wait for the run.completed webhook, then read the result and its trajectory.

What comes back

The answer is JSON in the session's result and in the run.completed webhook. The company's software shows it to the merchant as a checklist next to their catalog, each line linking to the record in Airtable, and holds back from publishing any product on the list until the merchant fixes it in their own base.

What the merchant checks: SKUs flagged as duplicates that are really two sizes of one product, which they rename; descriptions flagged as short that are fine for a simple item; and the missing_fields line, which usually means their base names a field differently from the template's guess. The session's accounts names the Airtable sign-in it used, and the transcript shows every tool call and the records it read, so a wrong flag can be traced to the row.

Run it on every week, per merchant

The company's scheduler sends one POST /o/{org}/sessions per connected merchant each week, before its own publishing job, with that merchant's agent id as agent and their base name in input, and a request_id that is a UUID v5 of the merchant's id and the week so a retried job does not start a second run. Runs past the key's concurrent limit are created queued and start as earlier ones finish.

A merchant can also press a "check my catalog" button, which sends the same request at once. Each answer arrives on the key's webhook, signed per Standard Webhooks; the handler matches execution_id to the merchant it saved at start.

The use case this job follows, with its automation and trigger: Run an agent per customer task.

Why run it on Rush?

Every merchant's base is shaped differently: one calls it "Price (USD)", another keeps photos in a field named "Pics", a third tracks stock per warehouse. A fixed integration needs a field mapping per merchant; the agent reads the schema each run and maps it, and says plainly when it could not. Each merchant's Airtable belongs to their own agent, so a run for one store never reaches another store's catalog.

The software company keeps one template. A change to the checks is one PATCH to the template, which reaches every merchant's agent on its next run, and GET /o/{org}/sessions?agent=<agent id> lists one merchant's runs and their total_cost_usd, so the company can see what each merchant's weekly check costs.

Frequently asked questions

Can the agent see all of a merchant's Airtable?

Only the bases and workspaces the merchant chose on Airtable's sign-in page. They can change that choice later in their Airtable account, and the next run sees the new set.

Will it fix the products for the merchant?

No. Airtable's MCP server has tools that update records, so the template's instructions and the prompt forbid them, and the run returns a list instead. The merchant fixes their own catalog, which keeps them the owner of what gets published.

What if a merchant disconnects Airtable?

Their agent's next run is refused 422 CONNECTION_NOT_READY before a sandbox starts, and GET /o/{org}/agents/{agent} lists the connection under missing_connections. The software can ask the merchant to connect again with the same sign-in call.

Does the merchant need a Rush account?

No. The merchant only signs in to Airtable. The agent, its runs and its bill belong to the software company's API key creator.

More Airtable jobs

More jobs like this: Run an agent per customer task

Keep reading