Give a Chat Agent Live Product Data

Connect an MCP scraper and compare live prices with an agent's first answer.

Introduction

30 Second Summary

Online prices move constantly. An assistant answering from memory can sound certain while quoting yesterday's number.

In this project, you will connect an assistant to Apify through MCP so it can fetch live product data. You will save answers to the same price question before and after the connection.

What You'll Build

Picture the second answer matching the live product page with a scraper run visible in the Apify console.

By the end of this project, you'll have:

A before-and-after comparison that shows exactly where an answer from memory was wrong or out of date.
A tool-enabled assistant with the scraper visible in its tool menu. Its answer matches the live product page.
An inspectable run record showing the fetched records and the credits used in the Apify console.
Secret Mission: An optional challenge to push your skills further.

Are there any prerequisites?

You'll need an Apify account with an API token. You'll also need an assistant that supports MCP servers.

You'll need a product page with prices that can change.

Before We Start

Live product data changes faster than an assistant's memory can keep up. A fair comparison starts with one fixed target.

In this step, you will prepare the two tools needed for the comparison. You will pair one live product page with one price question that stays unchanged.

In this step, get ready to:

Confirm that a web browser is available.
Confirm that your chosen assistant can accept an MCP server.
Choose one live product page and save an identical price question for both tests.

Confirm your browser and assistant

A web browser gives you the live product page for your comparison. Your assistant needs support for MCP servers so you can connect the scraping tool later.

What is MCP?

MCP gives an assistant a standard way to call external tools. In this project, that connection lets the assistant request current product data from a scraping tool.

Press Cmd+Space (macOS) or the Windows key (Windows) to open your search bar.
Type your browser's name into the search bar.
Press Enter to open the browser.
Go to the assistant you plan to use for this project.
Open the assistant's settings area.
Find the section for external tool connections or MCP servers.
Confirm that the assistant lets you add an MCP server.
Leave the assistant's tool connections unchanged so your first answer remains a clean baseline.

✔️ My assistant supports MCP

Good. Your assistant is ready for a tool connection later. Its baseline remains untouched for the first price question.

ⓧ I cannot add an MCP server

Review the assistant's official documentation to confirm whether it supports MCP servers.
Choose a different MCP-compatible assistant if support is absent.
Repeat the settings check in the new assistant.

Need help checking compatibility?

Assistant settings vary between products. Ask for help using the documentation for the assistant you selected.

Help me confirm whether my chosen assistant supports MCP servers.

Choose a live product page

Your product page becomes the live source for both sides of the comparison. A clear page shows one specific product with a current price that you can verify yourself.

Search the web for a physical product with a visible current price.
Open one public product page from the search results.
Confirm that the page shows a specific product name.
Confirm that the page shows a current price.
Confirm that the page loads without requiring a sign-in.
Record the product name here: your product name.
Record the page address here: your product page URL.

That is your live reference point secured. You can now judge every assistant answer against a price visible on the product page.

Lock in your comparison question

The before-and-after comparison only works when the prompt stays identical. Your question needs to request the current price for the specific product page you selected.

Use the saved product name and page address to write one current-price question.
Record the finished question here: your exact live-price question.

Your saved question becomes the source of truth for both runs. Every word and punctuation mark must stay the same.

Pause to predict whether your saved question identifies one product page clearly enough for both tests.
Read your saved comparison question once.
Confirm that the question asks for the current price.
Confirm that the question names your selected product.
Confirm that the question contains the saved page address.
Switch back to the product page from earlier.
Confirm that the product name matches your saved question.
Confirm that the current price is still visible.

You should see the same product represented in your saved question and on the live page. The visible price gives you a concrete value for checking each answer.

✔️ My comparison is ready

Your browser and MCP-compatible assistant are available. Your live product page and exact comparison question are also saved.

ⓧ I'd like to double check

Your comparison artifact should match this description:

One learner-selected, identical question requesting the current price of a product on a chosen live product page.

Your saved question is: your exact live-price question.

Your clean comparison setup is ready. Next up, you will prepare the scraping service and inspect the product data it returns.

Run Your First Product Scraper

Your before-and-after comparison needs a live source behind the second answer. A product page alone gives the assistant nothing it can call.

You will create an Apify account with a private API token ready for the later connection.

You will also run one Actor against the product page you chose. The resulting Dataset gives you the product record that later proves the tool used live data.

In this step, get ready to:

Create an Apify account.
Copy a private API token for later configuration.
Run one product Actor against your chosen page.

Create your Apify account and token

Apify Console manages your scraping runs plus their stored results. Your account gives you one place to launch the scraper and inspect its activity.

Visit the Apify sign-up page in your browser.

You should see the available account creation methods.

Choose one of the sign-up methods shown.
Complete the authentication flow for your chosen method.

You should land on the Dashboard in Apify Console. This confirms that your scraping account is ready.

Select Settings in the side menu.
Select API & Integrations in your account settings.

You should now see the controls for managing API tokens.

Copying a credential can feel risky, so a dedicated token keeps this project isolated for easier revocation.

Create a dedicated token in the API tokens area.

The new token is ready to copy for your later tool connection.

Copy the token value using its copy control.
Store the copied token in your password manager or another private credential vault.

Can't find your token?

Confirm that you selected Settings followed by API & Integrations.
Sign in again if Apify asks for a fresh session before showing security settings.
Help me find the API token controls without exposing my token.

Run a scraper for your product

Actors are cloud programs that perform specific scraping jobs. Each Actor documents the websites it supports plus the input it expects.

Select Apify Store in the side menu.
Search for a product scraping Actor built for the store on your chosen page.

You should see Actors designed for that store or product category.

Open an Actor whose description names your chosen store.
Read its README to confirm that it accepts individual product pages.
Return to the search results if the documented input does not support your page.

The selected Actor should describe a product output that includes price data.

Select the Input tab on the Actor page.

You should see the autogenerated form for this Actor's inputs.

Paste your chosen product page URL into the product URL input supplied by the Actor.

Starting the Actor consumes platform usage, with this first run limited to one page.

Before you start the run, which product details do you expect the Actor to return?

Click Start.

The run may stay in progress while the Actor gathers the page, so a short wait is normal.

Wait until the run shows a successful completion status.

Your first scraper run now has a completed console record.

Select the Output tab.
Confirm that a returned record matches the product on your chosen page.
Confirm that the returned data includes the product's current price.

That is the first live result locked in: your chosen page now has a structured product record.

No product record in Output?

Return to the Actor's README to confirm that it supports your chosen store.
Check that the product URL opens in a regular browser tab without requiring a sign-in.
Help me troubleshoot an empty product scraper output.

Inspect the run and credit usage

Each Actor execution creates a run record. The record ties its completion status to the output Dataset.

The usage breakdown shows the resources consumed by that run.

Before you open the latest run, do you expect its status, output, and usage to refer to the same scrape?

Select Runs in the side menu.
Open the most recent run for your product Actor.

The detail page should identify the Actor. It should correspond to the product-page run you just completed.

Confirm that the run status shows successful completion.
Locate the Usage label in the run details.
Select the question mark icon next to Usage to open the detailed breakdown.

You should see a usage value attached to this completed run. That value is the recorded credit usage for your scrape.

Usage missing from the run?

Return to Runs to confirm that you opened the completed product scrape.
Check the Usage column in the run list before reopening the run.
Help me locate usage for my completed Actor run.

Your live product record is now saved in Apify. Next, you will ask the identical price question with no tools connected to capture the baseline answer.

Save an Untooled Price Answer

Your first scraper run proved that the chosen product page can produce live product records. You now have a working source for the comparison ahead.

This step captures a clean baseline before your assistant gains access to the scraper. Saving the untooled response protects the evidence you will compare later.

In this step, get ready to:

Confirm that your assistant has no live-data tools enabled.
Ask the exact comparison question from earlier.
Save the complete response without changing it.

Confirm the assistant has no scraping tools

A useful baseline requires an assistant that cannot reach live product data. This keeps tool access as the single difference in the later comparison.

In your chosen MCP-capable assistant, start a blank conversation using the control for a new chat.
Open the tool settings for the blank conversation.
Disable every browsing or live-data tool for this chat.
Check the enabled-tools area to confirm that no scraping integration is listed.

Good. You now have a clean conversation that cannot call the product scraper.

Why Keep the Chat Tool-Free?

The later comparison needs one controlled change: access to the scraper. Using the same question isolates what the tool contributes.

Your API token has no role in this baseline. It stays in the private location from earlier.

Already See a Scraping Tool?

Disable the existing scraping integration before continuing.
Start another blank conversation after disabling the integration.
Confirm that the replacement conversation has no live-data tools enabled.
Help me confirm that my assistant has no live-data tools enabled.

Ask the unchanged price question

Changing the question can change how the assistant responds. The question from earlier must stay identical so the later result is comparable.

Before you send it, ask yourself whether the assistant can prove that its price came from the live product page.

Paste the comparison question you selected earlier into the blank conversation.
Send the question without changing its wording.
Wait for the assistant to finish its full response.
Preserve the first completed response as your baseline.

The assistant responds without a product scraper run behind it. That missing live-page evidence is the intended shortfall.

See Evidence of Browsing?

Return to the tool settings for this conversation.
Disable the browsing tool that produced the live-page evidence.
Repeat the unchanged question in a fresh conversation.
Help me create a tool-free baseline response.

Save the response verbatim

A verbatim copy preserves every detail in the answer. The exact wording becomes evidence in your final comparison.

Use the assistant's response-copy control to copy the entire answer.
Paste the copied answer into the text checkpoint below.
Preserve the answer exactly as the assistant wrote it.
Capture a screenshot that shows the comparison question plus the assistant's response.

Before you check the console, ask yourself whether this untooled chat should have triggered another scraper run.

Switch back to the scraping console from earlier.
Refresh the run history.
Confirm that no new run appears at the time of the assistant response.

You should still see the initial scraper run from earlier with no additional run tied to this chat. That gives you a clean untooled baseline for the comparison.

Did a New Scraper Run Appear?

Compare the new run's time with the time of your assistant message.
Disable the assistant's live-data integration if the times match.
Repeat the unchanged question in a fresh conversation.
Help me investigate an unexpected scraper run.

Your tool-free answer is saved exactly as the assistant produced it. Next, you will connect the scraper over MCP. Then the assistant can reach live product data.

Connect the Scraper MCP Server

Your assistant has already answered the live-price question from memory. That response is safely saved as the before side of your comparison.

The Model Context Protocol (MCP) gives an assistant a shared way to discover external tools. In this step, you will connect the Apify server with the private API token from earlier.

In this step, get ready to:

Add the hosted Apify MCP server to your assistant.
Authorize the connection with your saved API token.
Confirm the scraper tools in the assistant's tool menu.

Add the hosted MCP server

An MCP connection begins with a server address. This address tells your assistant where it can discover the scraping tools.

Open the official Apify MCP server page in your browser.
Find the MCP server URL section.
Copy https://mcp.apify.com.
Switch back to the MCP-compatible assistant from earlier.
Open the assistant's settings area for custom MCP servers or connectors.
Choose the option for adding a custom server.
Enter Apify as the connection name.
Paste https://mcp.apify.com into the server URL field.

What is an MCP server?

An MCP server publishes tools in a format the assistant understands. Apify uses this connection to expose scraping Actors that collect web data.

The server address provides the route to those tools. Your API token provides permission to use them through your account.

Authorize the server with your token

The server can advertise tools after receiving the correct account permission. Your API token supplies that permission.

This credential grants access to your Apify account. A private configuration screen is the right place for it.

My assistant has a token field

Use this path when the server setup provides a dedicated credential field.

Select the API token or Bearer token authentication option.
Paste the copied Apify API token into the credential field.
Save the MCP server configuration.
Confirm that Apify appears in the assistant's saved server list.

My assistant uses header fields

Use this path when the server setup asks for an HTTP header name plus its value.

Choose the option for adding an authentication header.
Enter Authorization as the header name.
Enter Bearer followed by one space in the header value field.
Paste the copied Apify API token immediately after that space.
Save the MCP server configuration.
Confirm that Apify appears in the assistant's saved server list.

My assistant uses JSON configuration

Use this path when the assistant manages MCP servers through a JSON configuration editor.

Open the assistant's MCP configuration editor.
Add the official connection object by pasting this configuration:

{
  "mcpServers": {
    "apify": {
      "url": "https://mcp.apify.com",
      "headers": {
        "Authorization": "Bearer <APIFY_TOKEN>"
      }
    }
  }
}

What does this configuration do?

The url value points the assistant to the hosted Apify MCP server.
The Authorization header sends your token as a Bearer credential.
The apify key gives the connection a recognizable name in the assistant.

Replace <APIFY_TOKEN> with the copied Apify API token.
Save the configuration.
Confirm that Apify appears in the assistant's saved server list.

Configuration not saving?

Check that every opening brace has a matching closing brace.
Confirm that <APIFY_TOKEN> is gone from the saved configuration.
Keep Bearer plus one space directly before your token.
Help me check my Apify MCP JSON configuration.

Keep the API token out of screenshots.
Keep the API token out of chat messages.

Connection not authorizing?

Check that the server URL is exactly https://mcp.apify.com.
Confirm that the saved token matches the private token copied earlier.
Check that the Apify server is enabled in the assistant's connection settings.
Help me troubleshoot my Apify MCP authorization.

Refresh the assistant's tool list

MCP clients often load available tools when a session starts. Refreshing the client forces it to read the saved server connection.

Restart the assistant if you use a desktop app.
Reload the assistant page if you use it in a browser.
Start a new conversation.
Enable the saved Apify connection for the conversation if your assistant uses per-chat tool controls.

Before you check, do you expect the tool menu to show the new scraper connection?

Open the assistant's tool menu.

You should see Apify scraper tools listed in the tool menu. This proves the assistant has discovered the configured MCP server.

Scraper tools still missing?

Confirm that Apify still appears in the assistant's saved server list.
Check that the Apify connection is enabled for the current conversation.
Restart the assistant again after saving the connection.
Help me find missing Apify MCP tools.

That's the bridge in place. Your assistant can now reach live scraping tools from a conversation.

The untooled answer is still saved. Next up, you will ask the same price question with Apify available. You can watch the assistant decide when to call the scraper.

Ask for the Live Price

Your assistant can already see its connected scraper tools. The next proof is a live tool call on the same question you asked earlier.

This controlled repeat keeps the wording fixed. The tool call through MCP becomes the change you are testing.

In this step, get ready to:

Submit the saved comparison question again without changing its wording.
Observe the assistant call a scraper through MCP.
Verify the returned price against the live product page.

Repeat the exact question

A fair before-and-after comparison needs one controlled input: the question. Reusing it verbatim keeps the tool connection as the change you are testing.

Switch back to the MCP-compatible assistant from the previous step.
Locate the comparison question you saved before connecting the scraper tools.
Copy the full question without changing its wording.

Before you submit it, do you expect the assistant to answer immediately or call one of its scraper tools first?

Paste the question into the assistant's message field.
Submit the question.
Watch the tool activity until the scraper finishes.

You'll see the assistant call a scraper through MCP. It then builds its response from the returned product records.

What changed this time?

MCP gives the assistant a route to the connected scraper. The scraper retrieves product records from the selected page for this request.

Those returned records supply the price used in the answer. The response now has evidence from the product page.

No scraper call?

Confirm that the scraper tools remain available in the current conversation.
Submit the saved comparison question again without edits.
Help me understand why the assistant is answering without using its scraper.

Verify and save the live answer

A tool call proves the connection works. A matching price proves the scraper returned the correct product record.

Wait for the assistant's response to finish.
Return to the chosen live product page from earlier.

Before you compare the two, do you think the live page will confirm the assistant's price?

Refresh the product page.
Locate the price for the exact product named in your question.
Compare the page price with the price in the assistant's answer.

You should see the same price in both places. That match confirms the assistant used the returned live product record.

Prices do not match?

Confirm that the product name in the answer matches the product page.
Match the product variant shown in the answer to the variant selected on the page.
Help me trace why the scraped price differs from the live product page.

Save the completed response verbatim as your tooled answer using the same method you used for the untooled answer.

What the tool call created

The assistant's request triggered a new product scraper run through MCP. That run stored the returned live product records in a new dataset.

The run and dataset are now ready for inspection in the scraping console. They form the evidence behind your saved tooled answer.

Your identical question now has a saved answer backed by a live scraper call. Next, you'll trace that answer to its run record, returned dataset, and credit usage.

Inspect the Scraper Run

Your assistant has answered the comparison question using prices from the live product page. A trustworthy answer still needs a trail that shows where its data came from.

The scraper run provides that trail. Its console record links the assistant's MCP tool call to returned product records.

In this step, get ready to:

Locate the scraper run triggered by the assistant through MCP.
Inspect the product records returned to the assistant.
Review the credits used by the run.

Find the MCP-triggered run

Each scraper call creates a run entry in the console. Its timing helps you distinguish the MCP-triggered run from your initial console test.

Switch back to the scraping console from earlier.
Use the console's recent run history to locate the newest completed scraper run.
Select the run created immediately after the assistant answered the comparison question.
Confirm the run target matches your chosen product page.

You should see a completed run whose timing matches the assistant's tool call. Its target should point to the product page used for your comparison question.

You have successfully connected the assistant's answer to one concrete scraper execution.

What does a run entry prove?

A run entry records one scraper execution. It gives you an audit trail between the assistant's request and the work performed in the scraping console.

Review the returned product records

The returned dataset is the source the assistant used for its tooled answer. Reading the records lets you trace a quoted price back to scraper output.

Switch from the run summary to the area containing its returned dataset.
Inspect the first returned product record.
Find a product named in the saved tooled answer.
Compare its recorded price with the matching price in the saved tooled answer.
Identify the fields that describe the product and its price.

You should see structured records containing live product details. At least one recorded product price should match the saved tooled answer.

You have now proved which scraper record supports the price in the tooled answer.

Why inspect the dataset?

The dataset is the evidence behind the assistant's response. Matching a returned record to the saved answer shows that the price came from scraper output.

Can't find a matching product?

Check that you selected the completed run created after the assistant called the scraper.
Search the returned records using the exact product name from the saved tooled answer.
Ask for help tracing the saved answer to the returned records.

Check the credits used

Credit usage shows the resources consumed by one scraper execution. Reviewing it makes the cost of the tool call visible.

Return to the summary for the same MCP-triggered run.
Locate the usage details attached to that run.

Before your final check, do you expect every piece of evidence to point to the same scraper execution?

Confirm the selected run shows a completed scraper execution.
Confirm its returned dataset contains the product record matched above.
Read the credits used by that same run.

You should see one completed run tied to the matched product record. You should also see a credit amount for that execution.

You have closed the audit trail from the assistant's tool call to the scraper output to its resource usage.

Your run now has a visible chain of evidence. Next up, you will place the two saved answers side by side to mark every stale or incorrect field.

Compare the Two Answers

The scraper run you inspected has already shown where the tooled answer came from. You now have the two saved responses needed to test whether live data changed the result.

A side-by-side comparison turns that test into visible evidence. Checking the untooled answer against the live records shows exactly what the MCP tool call corrected.

In this step, get ready to:

Place the saved responses beside the identical comparison question.
Check each field in the untooled answer against the live scraper records.
Save an annotated comparison showing what the tool call changed.

Build the side-by-side comparison

A fair comparison keeps the question identical. It also preserves each answer verbatim.

Return to the saved untooled answer from earlier.
Copy the full response verbatim.
Create a new document in a document editor.
Add Live Product Data Comparison as the document title.

You should see a blank comparison document with a clear title at the top.

Paste the exact comparison question beneath the title.
Insert a two-column table beneath the question.

You should see one question positioned above two empty columns.

Enter Untooled answer in the left column header.
Paste the copied untooled response beneath the left column header.

The left column should now preserve the complete response from before the scraper tools were connected.

Switch back to the saved tooled answer from earlier.
Copy the full response verbatim.
Enter With MCP in the right column header.
Paste the copied tooled response beneath the right column header.

You should now see both complete answers beneath the same question. Their side-by-side position makes every difference easier to trace.

Save the comparison document.

Audit the untooled answer

The returned records are the reference for current product data. Every concrete value in the untooled answer now gets checked against that evidence.

What counts as a field?

A field is one concrete product detail presented as current. The requested price is always a field in this comparison.

Treat another detail as a field when it appears in the saved answer.

Switch back to the returned product records from the MCP-triggered run in the scraping console.
Locate the exact product named in the comparison question.
Arrange the scraping console beside the comparison document.

You should be able to read the untooled answer beside the matching live product record.

Start with the first current product field stated in the untooled answer.
Compare its value with the matching field in the live record.
Highlight the untooled field when the values differ.
Add Wrong or out of date beside the highlighted field.
Add the matching live value beside the mark.
Repeat the check for every remaining field in the untooled answer.
Save the annotated comparison.

Your left column should now show every stale or incorrect field with a clear mark. Each mark should point to the value returned by the live scraper record.

Can't match a field?

Check that you selected the exact product named in the comparison question.
Mark a claim Unsupported by live record when the returned record contains no matching field.
Help me audit an assistant answer against a live product record.

Complete the comparison

A final sweep proves the comparison is complete. Reopening the document also confirms that your evidence stayed saved.

Before the final sweep, do you expect every current product claim in the untooled answer to have a clear result against the live record?

Read each current product field in the untooled answer one final time.
Confirm every mismatched value is highlighted.
Confirm every highlighted field includes the Wrong or out of date mark.
Confirm every marked field shows the matching live value.
Confirm the identical comparison question remains above both answers.
Save the completed comparison.

Close the comparison document.
Reopen the saved comparison from your document editor.

You should see the identical question above both complete answers. Every stale or incorrect field in the untooled answer should carry a mark with the matching live value.

You did the important part: every correction now points back to a live scraper record. Your saved comparison makes the value of the tool call visible in one place.

Secret mission

Audit the Answer's Evidence Trail

A live answer is useful. An auditable live answer is defensible. Trace each product claim from the saved comparison to its returned record. Then connect that record to the scraper run that produced it.

Clean Up Your Resources

Both scraper runs are complete, so no process is currently consuming credits. New requests through the MCP connection can launch more runs that use your monthly allowance.

Choose whether to keep your resources available, pause the assistant connection, or delete the project resources entirely.

Resources you used:

The configured Scraping MCP Server in your assistant.
The private API token used by the configured server.
The Initial Scraper Dataset from the first console run.
The MCP Product Dataset from the assistant-triggered run.
The saved Before-and-After Comparison.

Keep everything running

No action is needed. Choose this if you want to run more live product checks or preserve the comparison as evidence of what tool access changed.

Future tool calls can launch new scraper runs that use monthly credits.

Leave the configured Scraping MCP Server connected.
Keep the API token in its private storage location.
Preserve both scraper datasets in the scraping console.
Keep the saved Before-and-After Comparison for future reference.

Pause - I'll come back to this later

Disable the connection to stop new tool calls. Your datasets and saved comparison stay available for later.

In the MCP-enabled assistant's server settings, disable the configured Scraping MCP Server.
Confirm that the scraper tools are no longer available in the assistant's tool menu.
Keep the API token in its private storage location.
Return to the same server settings to re-enable the connection when you want to continue.

Delete - I don't want to use this again

Remove every project resource when you no longer need live scraper access.

Revoking the token is a firm cutoff. Your saved comparison stays intact until you remove it in the final cleanup action.

Remove the configured Scraping MCP Server in the assistant's server settings.
Confirm that the scraper tools no longer appear in the assistant's tool menu.
Revoke the API token used by the server from the account token settings in the scraping console.
Confirm that the API token is no longer active in the scraping console.
Delete the Initial Scraper Dataset from the dataset page linked to the Initial Product Scraper Run entry.
Delete the MCP Product Dataset from the dataset page linked to the MCP Product Scraper Run entry.
Use the completed run entries to confirm that neither dataset remains available.
Remove the saved Before-and-After Comparison from its original storage location.

Nice Work!

Strong finish! You gave a chat agent access to live product data through MCP. You proved the value of that tool call with a saved comparison that exposes every stale field in the untooled answer.

You've learned how to:

Run a product scraper against a chosen product page. Inspect the returned records in its dataset.
Connect a scraping MCP server to an assistant. Confirm that its scraper tools are available. Ask the identical price question again to receive an answer based on live records.
Trace the assistant's tool call through the scraping console. Review the matching run entry. Check the returned records. Inspect the credits used. Use that evidence to mark every wrong or outdated field in the untooled answer.
Secret Mission: Complete an optional project extension to push your live-data tool skills further.

Ready to quiz yourself?