Skip to content
← Live sellingMCP · Codex and other clients
SectionsExample promptsWhat your agent can doBefore you connectConnect CodexConnect PiConnect OpenCodeUse another agentStart with your products
Pi logoOpenCode logoClaude logoCodex logo

AI-powered live selling through MCP

Give Codex, Pi, OpenCode or another compatible agent access to your Shopify catalog, show planning, live controls and sales results through MCP.

Use with AI

Choose your assistant. We’ll copy the setup for you to paste.

Copy & open ChatGPTCopy & open Claude
For agents
Open MarkdownDownload guide
Agent index
Put your assistant to work

Start with a useful question.

  • Plan a show

    Help me pick products for my next live show and put them in a good order. Save the lineup as a draft.

  • Pick a sale format

    Which products in my draft would suit an auction, a fixed price or offers? Tell me why.

  • Check my prices

    Show me the starting prices and price limits in my draft. Is anything missing?

  • Get ready to launch

    Check my show before I publish it. What still needs fixing?

  • Schedule a show

    Help me schedule my next show for Friday evening. Check the date and my timezone with me first.

  • Catch up on the room

    What’s selling right now, what’s next, and are there any offers waiting?

  • Move to the next item

    Show me the next product and its price. Wait for my go-ahead before opening it.

  • Review offers

    Show me the offers waiting for a reply. Help me decide which to accept or counter.

  • Sell without going live

    Help me set up a standalone auction for one of my products. Ask me for the price and timing.

  • Let buyers make offers

    I want to take offers on a product. Help me choose it and save a listing draft.

  • Check payments

    What sold in my last show, and which purchases are still waiting for payment?

  • Plan the next sale

    Which products didn’t sell in my last show? Help me choose what to try again next time.

What your agent can do

Turn your catalog into a show

Ask your assistant to find Shopify products and variants, arrange the lineup and save a show draft. Your product selection becomes saved work in Stores on Live, ready for you to review and continue in the app.

Match the sale to the product

Let buyers compete in a rising-bid auction, wait for a falling price, buy at a fixed price or make an offer. Mix formats in one show and set each item’s starting price, reserve or price floor.

Give your show a place to sell

Check your draft for missing setup before you launch. Publish it as an event page on your Shopify storefront, or schedule its release. Shoppers get a place to watch, bid and buy from your store.

Keep the live sale moving

Ask your assistant to open the next product, close an auction or skip an item. Review incoming offers and accept, reject or counter them from the conversation. You present the products and start the camera on your device.

Keep selling between broadcasts

A product does not need a live show to sell. Create a standalone auction or an offer-based listing on your storefront. Set its pricing and schedule, then use your assistant to manage the offers that come in.

Know what needs your attention

Ask what sold, which purchases are paid and which still need payment. Review product outcomes and show attendance together, so you can see the result of the sale and decide what to follow up on.

Before you connect

You need a Stores on Live account and a connected Shopify store. Your agent must support remote MCP over Streamable HTTP and OAuth sign-in with PKCE.

OAuth lets you sign in to Stores on Live and grant access without putting your password in the agent’s settings.

Codex uses its published client identity. OpenCode uses the preconfigured client ID in the settings below. Neither needs registration with our team. Pi and other agents require a registered client ID and exact return address before sign-in. For registration help, contact us with the agent name, version and return address.

MCP server URL
https://kehk-systemic-production.up.railway.app/mcp

Connect Codex

Add Stores on Live
codex mcp add stores-on-live --url https://kehk-systemic-production.up.railway.app/mcp
Sign in
codex mcp login stores-on-live

Sign in, choose your store and approve access. No client ID or secret is needed.

Connect Pi

  1. Add the connection

    In a current Pi version, add the settings below to ~/.pi/agent/mcp.json. Keep any existing servers. Replace YOUR_REGISTERED_CLIENT_ID with the ID we supply.

    These settings use the return address http://127.0.0.1:8765/callback. Ask us to register this address with your client ID.

    Pi MCP configuration~/.pi/agent/mcp.json
    {
      "mcpServers": {
        "stores-on-live": {
          "url": "https://kehk-systemic-production.up.railway.app/mcp",
          "oauth": {
            "clientId": "YOUR_REGISTERED_CLIENT_ID",
            "callbackPort": 8765
          }
        }
      }
    }
  2. Sign in

    Pi sign-in command
    pi mcp login stores-on-live

    Sign in to Stores on Live. Select your store and review the access request.

  3. Check the connection

    Open Pi and use /mcp to check the server. Use /reload if Pi was open when you changed the file.

Pi’s MCP guide ↗

Connect OpenCode

  1. Add the connection

    Check opencode --version, then merge the matching settings into opencode.jsonc. Keep existing settings.

    No client secret is needed.

    OpenCode V2 configurationopencode.jsonc
    {
      "$schema": "https://opencode.ai/config.json",
      "mcp": {
        "servers": {
          "stores-on-live": {
            "type": "remote",
            "url": "https://kehk-systemic-production.up.railway.app/mcp",
            "protocol": "auto",
            "oauth": {
              "client_id": "stores-on-live-opencode",
              "callback_port": 8765,
              "redirect_uri": "http://127.0.0.1:8765/callback"
            }
          }
        }
      }
    }
    Using OpenCode 1.x?

    For 1.18.32, use these settings instead.

    OpenCode 1.x configurationopencode.jsonc
    {
      "$schema": "https://opencode.ai/config.json",
      "mcp": {
        "stores-on-live": {
          "type": "remote",
          "url": "https://kehk-systemic-production.up.railway.app/mcp",
          "oauth": {
            "clientId": "stores-on-live-opencode",
            "callbackPort": 8765,
            "redirectUri": "http://127.0.0.1:8765/callback"
          }
        }
      }
    }
  2. Sign in

    OpenCode sign-in command
    opencode mcp auth stores-on-live

    Sign in to Stores on Live. Select your store and review the access request.

  3. Check the connection

    Run opencode mcp list. Check that Stores on Live is connected.

OpenCode’s MCP guide ↗

Use another agent

Add a remote MCP server using the URL above. Use the client ID and return address registered for that agent. Keep OAuth enabled, then sign in.

A local-only MCP client cannot connect to this URL on its own. Contact us before you set up a different agent so we can check its sign-in support.

Start with your products

Try asking
Show five products from my connected store. Do not change anything.

You should receive products from the store you approved. Agents with text-only results can use the tools. An agent must also support MCP Apps to show the selling panels inside the conversation.

For ‘client not registered’ or a rejected callback, check that you used the settings for your client and version. For expired or revoked access, sign in again from the assistant. If the connection still shows old tools or sign-in settings, refresh its server metadata where supported and start a new chat. For help, contact support with the client name, version and error text; do not send passwords, tokens or authorization codes. Contact support.

Open Agent connections in Stores on Live to review or revoke your assistant’s access. To approve additional permissions, reconnect from your assistant and review the new access request. Refreshing a page or connection metadata does not grant more access.

Contact usPrivacyTerms

Product names and logos belong to their owners. OpenAI and ChatGPT marks belong to OpenAI.