Introduction
Rolando (Host):Welcome, everyone. Today's session is a live walkthrough — a little different from a traditional slide-heavy webinar. Kenton Snyder, our Product Manager at Intentwise, will show us how Claude Live Artifacts can be used to build custom Amazon dashboards: what the workflow looks like, what's possible today, and where teams may still need the right data layer behind the scenes.
Kenton Snyder:Thanks, Rolando. If you have a Claude subscription, I'd encourage you to open Claude desktop — we'll be working in Claude Cowork today. Even if you're on a different AI tool, there'll be plenty of valuable takeaways for you. I'll also be connecting data through the Intentwise AI Gateway, our MCP solution that pipes Amazon data into tools like Claude.
What Are Claude Artifacts and Live Artifacts?
Kenton:Artifacts have been part of Claude since its early days — they're HTML files that Claude generates through Claude Cowork. Claude Cowork is a wrapper built on top of Claude Code that makes it much easier to write code files without needing to be technical. It's particularly strong at writing HTML, which is the foundation for most artifacts.
The newer feature — Live Artifacts — is what really unlocks the value here. Live Artifacts take those same HTML files and turn them into self-refreshing dashboards. By connecting to an MCP server like Intentwise AI Gateway, the artifact pulls fresh data every single time you open it. The refresh typically takes 15 to 30 seconds, depending on data complexity. No manual exports, no rebuilding — just always-on dashboards.
What Can You Build?
Kenton:Here are a few examples I built over the last couple of days:
Daily Operations Hub — A top-line Amazon performance view showing ad spend, ROAS, total orders, daily trend lines with comparison periods, top- and bottom-performing campaigns, and anomaly flags. Fully filterable by date range and ad type.
Vendor P&L — Shipped revenue, ad spend, revenue after ads, ordered revenue — all in one view. You can layer in FBA fees and coupon data too.
Inventory Overview — Sellable units, open POs, and ASIN-level breakdowns at a glance.
All of these are built using plain-language prompts. No code required.
Building a Live Artifact: Step-by-Step
Kenton:Here's the exact workflow to build a new live artifact in Claude Cowork:
Step 1 — Define your style. You can be as loose or as specific as you want. Options include uploading a screenshot of a dashboard you like, naming a design system (e.g., "modern SaaS style using Inter fonts and red accent colors"), or providing exact hex codes and font names. If you're an agency sharing dashboards with clients, giving stricter guidelines ensures brand consistency.
Step 2 — Describe your data and metrics. You can keep it high-level ("create an inventory overview") or be explicit. For this demo, the prompt was:
"Build a live artifact of top-line vendor performance. Use the last 30 days of data compared to the previous 30 days. Start with KPI scorecards showing current vs. previous values for total vendor sales, ad spend, ad sales, ROAS, and total vendor orders. Add a daily trendline for total vendor sales and ad spend. Finally, include a table breakdown by ASIN showing total sales, total orders, glance views, and conversion rate."
Step 3 — Let Claude connect to your MCP and build. Claude will ask permission to call tools from your connected MCP server. It will locate your account, identify the right data fields, pull the data, and write the HTML artifact. This typically takes a couple of minutes for the first build.
Step 4 — Troubleshoot if needed (usually a one-time step). If the data connection isn't fully established on the first render, just tell Claude: "The data is not fully connected — check the connection between the data and the visualization." Once fixed, the same code runs every refresh — you won't need to troubleshoot again.
Output: KPI scorecards, a daily trend chart, and an ASIN-level performance table — all live, all accurate, all generated without writing a single line of code.
Managing and Sharing Live Artifacts
Kenton:Once your artifact is live, here's what you can do with it:
Refresh: Every time you open the artifact, it auto-refreshes with the latest data. You can also force a refresh using the Reload button in the top right.
Version history: Past versions of the artifact are saved automatically. You can go back and compare what your data looked like 2 or 3 days ago.
Export/share: There's no live hosted link feature yet, but you can download the current state as a PDF and share it via email, Slack, or any other channel.
Edit: Click the Edit button to reopen the chat and modify the artifact — add filters, change metrics, adjust the layout, or fix a formula.
Connecting an MCP Server to Claude Cowork
Kenton:To connect a new MCP server: go to your profile icon → Settings → Connectors. You'll see all currently connected integrations. Click "Add Connector" to browse built-in options (like BigQuery, Slack, Figma) or set up a custom connector. You can also toggle connectors on or off per individual chat session using the "+" button in the chat interface.
What Data Is Available Through the Intentwise MCP?
Kenton:The Intentwise AI Gateway MCP covers a broad scope of Amazon data, including:
- Advertising data (Sponsored Products, Sponsored Brands, Sponsored Display)
- Total sales across Seller Central and Vendor Central
- Inventory and FBA reports
- Finance reports
- Coupon data
- Search Query Performance (SQP)
Walmart and Meta data sources are on the roadmap. AMC data is not currently available through the MCP but is in development.
Four Things to Evaluate in Any MCP Server
Kenton:As you explore MCP options for live artifacts, keep these four criteria in mind:
1. Data completeness. Make sure the MCP covers all the datasets you need to analyze — ads, finance, inventory, etc. — so Claude has the full picture.
2. Data accuracy. AI is prone to hallucinate. Strong MCP servers return consistent results every time they're called. Ask providers for accuracy metrics and check for consistency across refreshes to minimize errors in your dashboards.
3. Cost optimization. Every MCP call uses tokens. Good MCP servers limit the number of rows returned per request so they don't overload Claude's context window and drive up your costs unnecessarily.
4. Data security. Look for MCP servers built in contained environments — platforms like AWS Bedrock or Microsoft Azure can run LLMs locally without sending your data to external servers or allowing it to be retrained on. Always ask your MCP provider how data isolation is handled.
Q&A Highlights
Q: Do we need to set up a scheduled task to refresh the artifact daily?A: No. Live Artifacts refresh automatically every time you open them. No scheduling needed.
Q: Can we filter by a specific ASIN?A: Yes. Just describe the filter in your prompt — e.g., "add a filter to view data by ASIN" — and Claude will build it in.
Q: How do we fix an incorrect metric?A: If a metric looks wrong, first check whether it's a data quality issue from the MCP (ask Claude which field it's using). If it's a calculation error in the artifact itself, open the chat and tell Claude the correct formula. It will debug and fix it.
Q: Can you turn an existing artifact into a live artifact?A: Yes. Share the file with Claude in Claude Cowork and prompt: "Can you turn this into a live artifact?" It will handle the rest.
Q: Is there a difference between Sonnet in normal vs. adaptive mode for this use case?A: Minimal difference for most artifact builds. Adaptive thinking helps more for open-ended research or complex design decisions. For explicit, metric-driven builds like these, standard mode works well.
Q: Will live artifacts replace the Intentwise platform dashboard?A: Not in the immediate term — but the possibilities for custom dashboarding are genuinely open-ended and will continue to evolve as teams get more comfortable with Claude Cowork.
Closing
Kenton:Thanks to everyone who joined today. The goal was to show you that building custom, live Amazon dashboards is genuinely accessible — no code, no complex setup, just a clear prompt and a solid MCP connection. As your teams get more comfortable with this workflow, the customization possibilities are really only limited by what you want to see.
If you'd like to explore Intentwise AI Gateway, we're offering a 2-week free trial — reach out at kenton@intentwise.com or fill out the form linked in the session. And join us for next week's webinar: "Your Amazon Data is AI Ready — Now What?"
