Anthropic shipped Live Artifacts to Claude Cowork between April 21 (the LinkedIn announcement) and April 28 (the formal help-center publication). Eight days from “shadow rollout” to “documented feature,” and in that window people on X already built dashboards that replaced custom Python scripts, Tableau workspaces, and Notion AI workflows — most in well under an hour.
Live Artifacts are the answer to the most common Claude complaint: “I built this great dashboard, then closed the chat, and now it’s gone.” A Live Artifact is a persistent, interactive HTML page — a dashboard, tracker, comparison tool, status board — that pulls fresh data from your connected apps every time you open it. Saved to a “Live artifacts” tab inside Claude Cowork on Desktop. Version history. No backend. No hosting. No monthly fee for the artifact itself (though Cowork still requires a paid Claude plan).
The Anthropic help doc is correct, and incomplete. It tells you the feature exists. It doesn’t tell you what to actually build first, what limitations you’ll hit, or which prompt patterns get you to a working dashboard fastest. Five templates below, each takes about 15 minutes once your connectors are set up, each is grounded in a real build that real people on X are already running. Prompts included.
Before You Start: Setup (5 Minutes, One Time)
What you need:
- A paid Claude plan: Pro, Max, Team, or Enterprise. The free tier doesn’t include Cowork.
- Claude Desktop on macOS or Windows. Live Artifacts are a Desktop feature.
- At least one connector active. The default integrated connectors include Asana, Linear, Slack, calendar apps, Google Drive, Gmail, Notion, Salesforce, HubSpot, Stripe, and Jira. You can browse the full list at the Claude connector directory.
To create a Live Artifact two ways:
- From a Cowork chat: just ask, e.g., “Build me a dashboard that shows open tasks by project, pulling from Asana and Linear.” The artifact saves automatically.
- From the Live artifacts tab: open Cowork → “Live artifacts” in the sidebar → “New artifact” (top right) → “Chat with Claude.”
For the more advanced templates below (specifically the trading-positions one), you’ll need a custom MCP server. Anthropic’s official guide to remote MCP connectors walks through the Settings → Connectors → Add custom connector flow. If you’re on Team or Enterprise, your org admin needs to add custom connectors at the Organization-settings level first.
That’s it. Now five real templates.
Template 1: The Wheel-Strategy Options Tracker (Jason Luongo’s Build, 720+ likes)
Why this one is the killer use case: Live data + persistent UI + zero backend. The trader who built it (@JasonL_Capital, April 21–22) replaced what would have been a custom Python script + database + frontend with one Claude artifact and a Notion table.
What it does: Tracks open options positions in a “wheel strategy” (selling cash-secured puts → assigned shares → selling covered calls). Pulls live option Greeks (delta, theta, premium) every time you open it. Shows total portfolio delta, daily theta, monthly premium, and annualized return on capital. Roll and Close buttons inline.
Connectors needed: Notion (positions storage) + a brokerage MCP. Public.com publishes a verified MCP server for Claude Desktop for portfolio review and trade execution (US-only currently; Robinhood and Interactive Brokers haven’t published verified MCPs yet — confirm if your broker supports MCP before starting).
The exact 4-prompt sequence (Jason’s, lightly cleaned):
Prompt 1: "Set up a Notion database under my Wheel Tracker page for
logging my wheel strategy trades. Columns: ticker, position type
(CSP / shares / CC), strike, expiration, premium received, current
price, delta, theta, status (open / rolled / closed)."
Prompt 2: "Build a live page called wheel-tracker. It should read my
open positions from this Notion database and pull live option Greeks
from Public.com via the MCP connector. Refresh on open."
Prompt 3: "Add a view grouped by days to expiration with Roll and
Close buttons next to each position. At the top show total portfolio
delta, daily theta accrual, and monthly premium collected."
Prompt 4: "Add charts showing which phase of the wheel I'm in for
each position, monthly premium history, and annualized return on
capital. Use a card layout, not a wall of tables."
What works: Auto-refresh on open, full version history, persistent across sessions and weeks.
What to know: Mid-session refresh is manual (click the refresh button) — Greeks go stale if you keep the artifact open all day. Public.com MCP is US-only. Use this template only if your broker has an MCP server or you’re willing to use a generic stock-quote API via a custom MCP.
Template 2: Weekly OKR Dashboard from Notion (@TestingCatalog, 472 likes)
Why this template: OKRs are the most universally requested dashboard in tech, and Notion is where most teams already track them. Result: zero data migration. The dashboard reads what’s already there.
What it does: Pulls active OKRs from a Notion database, calculates current progress against targets, flags OKRs at-risk (less than expected progress for the time remaining), and shows trend lines for the previous 6 weeks.
Connectors needed: Notion only.
The 3-prompt sequence:
Prompt 1: "Read my OKRs database in Notion (filter: status =
'Active', current quarter). For each OKR, list the objective, the
key result, the target value, the current value, the % progress,
and the days remaining in the quarter."
Prompt 2: "Build a live artifact called weekly-okrs. Show me a
card per OKR. Color each card by status: green if on track (% time
elapsed ≤ % progress), yellow if within 10% behind, red if more.
At the top, give me a single sentence: 'On track to land X of Y
key results this quarter.'"
Prompt 3: "Add a 6-week trend line per OKR using the historical
progress entries in the database. If the trend is flat or
declining, mark the card as stalled regardless of overall %."
What works: Updates automatically every time you open it Monday morning. The sentence at the top is the part that earns the artifact a place in your workflow — it’s a single-line read of the entire quarter.
What to know: If your team logs OKR progress sporadically, the trend line will be noisy. Either backfill historical entries or skip Prompt 3.
Template 3: SEO + AI Search Visibility Dashboard (SE Ranking’s Build)
Why this template: The fastest-growing dashboard category in 2026 isn’t traditional analytics — it’s “how does my brand appear in AI answers?” SE Ranking (April 27 X post) shipped a Live Artifact that combines Google Analytics MCP with their own MCP to track brand mentions across ChatGPT, Gemini, Perplexity, and Google AI Overviews. Same principle works without SE Ranking — anyone can build the GA4 half today.
What it does: Pulls top organic pages, click-through trends, AI Overview impressions (when GA4 has them), and traffic deltas vs. competitors. Optional: brand mentions in AI answers if you have an SE Ranking, Profound, or AthenaHQ MCP.
Connectors needed: Google Analytics MCP (Anthropic publishes one) + optionally an AI-visibility MCP.
The prompt:
"Read my GA4 data via the Analytics MCP. Build a live artifact
called ai-visibility-weekly that shows: (1) top 20 landing pages
by organic clicks last 7 days, (2) week-over-week change in
impressions, (3) impressions in AI Overviews specifically (filter
search appearance = 'aio'), (4) bounce rate and time-on-page for
the top 20 pages, (5) any landing page with >50% week-over-week
traffic drop flagged in red. Refresh on open. Card layout, not a
table."
What works: GA4’s AI Overview impressions data is the closest thing to a single-page reality check on whether your content is being surfaced by Google’s AI summaries. Having it auto-refresh weekly turns an “I should check that” into a “I see it the moment I open Cowork.”
What to know: GA4’s search appearance = 'aio' dimension is still rolling out; if you don’t see it, ask Claude to fall back to organic clicks delta only. Brand-in-AI-answers tracking requires a third-party MCP — not all SEO platforms have shipped one yet.
Template 4: The Daily Morning Business OS
Why this template: This is the one that replaces your “ten tabs open at 9am” routine. Multiple builds on X this week (@DailyAIOptimist on April 28, others) follow this pattern: one artifact, one open at the start of the day, every important thing in front of you.
What it does: Combines your calendar (today’s meetings + the next 3 days), unread Gmail (filtered to actionable), Slack mentions in the last 12 hours, and active Notion tasks marked “Today.” Shows a one-paragraph “what to focus on” summary at the top.
Connectors needed: Google Calendar, Gmail, Slack, Notion (any subset works).
The prompt:
"Build a live artifact called morning-os. At the top, give me a
two-sentence summary of my day: what's the most important thing
on my calendar, and the one email/Slack thread I should respond
to first.
Below that, show:
- Today's meetings (calendar) with the Zoom link extracted
- Unread Gmail with 'response needed' in subject or body, last
18 hours
- Slack DMs and @-mentions, last 12 hours, in any channel
- Notion tasks with status = 'Today'
Refresh on open. If something is empty, just hide that section,
don't show 'No items'."
What works: The two-sentence summary is the part that earns this artifact a daily open. Without it, you’re just looking at four lists. With it, Claude is doing actual triage.
What to know: First time you build it, expect to spend 5-10 minutes refining the Gmail filter. “Response needed” misses some real ones and catches some marketing emails. Iterate on the prompt until the catches match what you’d flag manually.
Template 5: Live Client Report (@JulianGoldieSEO, Hot-Leads Variant)
Why this template: Agency owners and freelancers send the same recurring client report week after week. Live Artifacts collapse the prep time from “two hours every Friday” to “open the artifact, screenshot, send.”
What it does: Pulls a single client’s KPIs (traffic, leads, top pages, conversion events) from GA4 + your CRM, formats them as a one-page report, and updates on open. You ship one screenshot every Friday.
Connectors needed: Google Analytics MCP + your CRM (HubSpot or Salesforce MCP available; or a Notion/Google Sheet for smaller agencies).
The 2-prompt sequence:
Prompt 1: "Read GA4 for the property [client-property-id] and read
HubSpot for the deal pipeline tagged [client-tag]. Show me what
data is available for the last 30 days."
Prompt 2: "Build a live artifact called [client-name]-weekly with
the following sections: (1) Headline metric: organic traffic vs
previous 30 days. (2) Top 10 landing pages by sessions, with the
delta. (3) Pipeline summary: deals created, deals advanced stage,
deals closed-won, total $ value, last 30 days. (4) Hot leads:
HubSpot contacts with lead score > 70 and last activity in last
7 days, with their next-best-action drafted. (5) Three-bullet
'what to recommend' section based on the patterns above. Use the
client's brand color [#hex] in headings."
What works: Once built, it’s the fastest weekly client check-in you can run. Open the artifact, screenshot, paste into the email, send. Done.
What to know: “Next-best-action drafted” requires Claude to make a recommendation it might be wrong about — review before sending. The brand-color customization is what makes the screenshot look like yours, not a Claude template.
What Anthropic’s Help Doc Won’t Tell You
Five honest limitations from real-user signal across X this week:
Refresh is on open, not push. Live Artifacts pull fresh data when you open them. They don’t auto-update if you leave them open for hours. The trading-positions template needs manual refresh mid-day; same for any artifact watching live-changing data.
Local storage only. Live Artifacts live on your Mac or Windows machine. They aren’t synced across devices and aren’t shareable at launch. If you build something great, you can’t (yet) send your colleague a link to your artifact.
MCP setup is technical for custom servers. Built-in connectors (Notion, Slack, etc.) work via OAuth in two clicks. Adding a custom MCP server (like Public.com’s brokerage one) is a JSON-config + Homebrew/UV install — fine for devs, friction for everyone else.
Paid plans only. Pro ($17/mo annual or $20/mo monthly), Max (3 usage tiers starting around $100/mo), Team ($20/seat/mo annual or $25 monthly), or Enterprise (custom). No Live Artifacts on the free tier. See anthropic.com/pricing for the current numbers.
You’re still dependent on Claude for hosting. There’s no “export to standalone HTML” that keeps the connector links live. If Anthropic’s prices change or the feature changes shape, your dashboard goes with it. For mission-critical workflows, treat it as a productivity layer, not a system of record.
The framing in the wild — “Anthropic is going after Bloomberg Terminal” — overstates what Live Artifacts is today and understates what the trajectory looks like. As of April 28, this is a productivity feature for individuals and small teams. The Tableau-replacement claims are aspirational. But the moat for those tools — “we built the UI, you can only use what we built” — just got noticeably weaker. As @testingcatalog put it: “GenUI is here. Value moves from UI interfaces to whoever lets you build your own UI.”
What to Build First
If you’re new to Live Artifacts, start with Template 4 (the Morning Business OS). It uses the most common connectors most people already have approved (Calendar + Gmail + Slack + Notion), the prompt is forgiving, and you’ll feel the value the first morning you open it.
If you’re a developer or trader, Template 1 is more interesting because it forces you to add a custom MCP. Once you’ve done that once, every other custom-data dashboard becomes available.
For agency owners and consultants, Template 5 has the highest weekly time savings.
For SEO and content people in 2026, Template 3 is the dashboard you didn’t know you needed. AI Overview impressions are the metric that will matter most this year, and most analytics setups don’t surface them clearly.
For anyone running a team, Template 2 is what your Monday already needs.
Cross-link: if Live Artifacts is your first Claude Cowork feature, the Cowork beginner’s guide covers the surrounding workspace. If you’re evaluating Cowork against Microsoft and Google’s competing assistants, see our Microsoft Agent 365 4-models analysis for the IT-buyer side of the same shift.
