Articles in this section

Getting Started with the CommentSold MCP Connector

The CommentSold MCP (Model Context Protocol) Connector allows you to access your shop data and settings directly from your AI LLM tool of choice.  Once connected, you can ask your LLM questions about your products, inventory, orders, and collections, and take actions in your store using plain conversational language.

Use the table of contents below to navigate to specific sections of this article:

Supported AI LLM Tools


  • Claude.ai
  • ChatGPT
  • Other LLM tools that support MCP custom connectors

Requirements


Before connecting, make sure you have:

  • A LLM account (EX. Claude.ai)
  • Active access to your CommentSold Admin panel
  • The MCP feature is automatically enabled for shops on an AI plan (contact support if you would like to discuss plan options)
  • Associated permissions enabled for your account within CS Admin (see below)

Connecting The CommentSold Connector (Claude.AI example)


 

  1. Open Claude.ai and navigate to Customize > Connectors.
  2. Click Add > Add custom connector
  3. Name the custom connector CommentSold and insert https://mcp.commentsold.com/mcp in the URL field
  4. Click Connect
  5. You'll be prompted to log in with the same username and password you use to access your CommentSold Admin panel
  6. Authorize the connection. Claude will confirm the connector is active.

Connecting The CommentSold Connector (ChatGPT example)


 

  1. Open ChatGPT and navigate to Settings > Security and login.
  2. Enable Developer mode.
  3. Navigate to Plugins and click the + to add a new plugin.
  4. Name the custom connector CommentSold and insert https://mcp.commentsold.com/mcp in the URL field.
  5. Click Create.
  6. Sign in with CommentSold. If you’re already logged into CommentSold on your browser, this part will happen automatically after clicking to sign in.

Once connected, you can start a new conversation and reference your store data directly.

What You Can Do With The Connector


The connector gives your LLM access to your store across five main areas. Below is a breakdown of what's available.

Products

What you can ask your LLM to do:

  • Search your product catalog by name, SKU, status, or other attributes
  • Look up per-variant inventory for any product
  • View inventory history for a specific product or variant
  • Archive or unarchive a product
  • Update a variant's price

Example prompts:

  • "Search for all active products with 'floral' in the name."
  • "What's the current inventory on the blue size medium of SKU FL-1042?"
  • "Archive the product 'Spring Tote — Sold Out.'"
  • "Update the sale price on variant 8821 to $24.99."

Guardrails:

  •   Free-gift products cannot be priced or archived through the connector.
  •   Inventory adjustments are capped at ±10,000 units per action.
  •   Dropship products have built-in protections.

Collections

What you can ask your LLM to do:

  • List all collections in your store
  • View which products belong to a specific collection (including drafts)
  • Create a new collection
  • Add or remove products from a collection (in bulk)
  • Delete a collection

Example prompts:

  • "List all my collections and how many products are in each."
  • "Add product IDs 1020, 1021, and 1022 to the Summer Sale collection."
  • "Create a new collection called 'August Arrivals.'"
  • "Delete the 'Holiday 2024' collection."

Guardrails:

  •   Smart collections cannot be created or modified through the connector.
  •   Deleting a collection will not delete the products in it — products remain in your catalog.
  •   Claude will ask you to confirm before deleting a collection.

Orders

What you can ask your LLM to do:

  • Look up a specific order by ID
  • List recent orders with filters (status, date, etc.)

Example prompts:

  • "Show me all orders from the last 7 days."
  • "Pull up order #48821."

Sales & Shop Data

What you can ask your LLM to do:

  • Get a revenue summary by time period (yesterday, this week, last week, this month, last month)
  • View your shop's current settings and payment methods configuration

Example prompts:

  • "What were my sales this week vs. last week?"
  • "Show me my current payment method settings."

Feedback

You can submit feedback or report issues directly from a LLM conversation. Your LLM will log your feedback and route it to the CommentSold team.

Example prompt:

  • "Submit feedback: the inventory history tool isn't returning results for archived products."

Permissions Reference


The connector uses your existing CommentSold Admin permissions. The actions available to your LLM are limited to what your account is authorized to do in the admin panel.

Action Category Required Permission
View/Search products View Products
Edit products/pricing Manage Products
View orders Manage Orders
Mark order as shipped Manage Orders
View sales summary View Dashboard
View shop settings Manage Payments
Submit feedback No permission needed

Expected Behaviors and Guardrails


  • Your LLM can only access data and take actions that the connector tools explicitly support. It cannot browse your admin panel, access files, or take actions outside of the listed capabilities.
  • Actions that are irreversible or customer-facing always require human approval before they are executed.
  • Your LLM will not perform bulk price changes, issue refunds, apply store credit, or run data exports through the connector. These are not in scope for the current version.
  • If your LLM can't find something or a tool returns an error, it will tell you and suggest next steps.

Troubleshooting


Here are some common errors and their resolutions:

Error: Connector not showing up in your LLM

The MCP feature is automatically enabled for shops on an AI plan (contact support if you would like to discuss plan options).

Error: Login not working

Make sure you're using your CommentSold Admin credentials, not a personal LLM login or a separate app login.

Error: Your LLM says it doesn't have access to a feature

The action may require a permission your account doesn't have, or the feature may not be available in the current connector version.

Was this article helpful?
0 out of 0 found this helpful