twimco.Let’s talk ↗︎
Menu
Field notes / Reviews

Stripe agent analytics review: worth connecting for read-only finance questions first

Stripe now lets agents query analytics through MCP, Queries and Reports for every account, with Queries free through January 31, 2027. The useful review is which questions you'd trust after you open the SQL.

Editorial engraving of a sunny analytics nook where a controller inspects soft query lines on a laptop while an analyst holds a blank printed chart.

Verdict: Stripe's November 5, 2026 agent analytics launch is worth connecting in read-only mode for teams that already use Claude or Cursor with Stripe MCP, because agents can run Stripe-defined metrics, subscription templates and financial reports against a modeled schema of 300+ tables, and you can inspect the SQL behind each answer. Keep write tools and broad secret keys out of the same session. Queries API is free through January 31, 2027, then usage-based from February 1; Reports API stays free. Do not use it to invent MRR by summing invoices yourself when stripe_analytics already defines the metric.

The verdict

On November 5, 2026 Stripe announced that compatible agents—Claude and Cursor are named—can analyze Stripe data through the Stripe MCP or directly through the Queries API and Reports API. The capability is available to all Stripe users. Agents can query Stripe-defined metrics such as MRR, run subscription and billing templates, execute SQL against reporting tables, and generate financial reports. For each analysis you can inspect the underlying SQL, confirm how the result was calculated, and edit the query.

Here's our verdict:

Worth connecting for read-only finance questions if your team already lives in an agent that can use Stripe MCP and you want MRR, payouts, disputes and churn explained without exporting CSVs by hand.

Inspect the SQL every time it matters. If you would not paste the SQL into a finance review, do not paste the agent's answer into the board deck.

Separate this from write access. Stripe's broader MCP still has write tools and key rules. This review is about analytics scope. Pair it with your agent-key migration work; do not dump a full-access secret into a chat that also runs SQL.

Watch the free window. The Queries API is free through January 31, 2027. Usage-based pricing begins February 1, 2027, with no monthly subscription required. The Reports API remains free.

Review by the SQL you'd inspect

Illustrative prompts below follow Stripe's published examples. The discipline is what you do after the answer lands.

Reconcile payouts. Ask the agent to tie three months of payouts to payments, refunds, disputes, fees and transfers, and to show amounts that do not tie out. Open the SQL. If the joins skip a transfer type you use, fix the query before you trust the exception list.

MRR movement. Ask for MRR, expansion, contraction, churn and reactivation by product. Prefer Stripe-defined metrics via stripe_analytics over an agent that "sums invoices." Stripe's docs warn that DIY MRR will not match Stripe's definition.

Failed payments. Ask for failures that did not succeed within three days, grouped by decline reason and method. Check that the date window and retry logic match how your billing team already reports.

Fraud segments. Ask for chargeback rates by segment. Keep alerting hooks human-reviewed at first—auto-Slack on a bad query is how finance loses a week.

Open invoices. Age open and past-due invoices by customer. Confirm the agent is not mixing test-mode rows into a live narrative.

Decision guide

Read the SQL before you trust the slide

ConsiderationAgent outputBefore you ship it
MRR bridgeNarrative of expansion/churn.Confirm Stripe-defined metric, not DIY sum
Payout reconcileList of untied amounts.Inspect joins for fees and transfers
Fraud hotspotsSegments with rising disputes.Human review before auto-alerts
Open invoicesAged AR table.Exclude test mode; match billing defs
Decision framework: treat agent analytics as draft until SQL is reviewed.
If you would not paste the SQL into a finance review, do not paste the agent's answer into the board deck.

Where it falls short

Own account only. Connect platforms cannot use these tools to read connected-account data. If your world is a platform, this launch does not solve connected-account analytics.

Preview surface. Metrics and reports can change during public preview. Pin the queries you rely on and re-check after Stripe updates.

Agent math drift. If the agent calculates a metric from raw rows instead of calling Stripe-defined metrics, numbers can diverge. Prefer stripe_analytics for official definitions.

Not a substitute for controls. SQL you can inspect is necessary, not sufficient. Still decide who may run Queries after the free window, which workspaces get OAuth scopes, and whether Results may leave the company laptop.

Different problem from agent keys. Migrating MCP away from full-access secrets by November 30 is a separate deadline. Solve keys and analytics as two tickets.

Pricing, as published

From Stripe's Nov 5 blog and Analyze with AI docs as of November 6, 2026:

  • Queries API: free through January 31, 2027; usage-based credits from February 1, 2027; no monthly subscription required
  • Reports API: remains free
  • OAuth scopes called out for metrics read and financial reports read

Budget a February cost conversation now if Queries volume will be high. Do not assume "AI analytics" stays free forever.

Who should skip it

Skip if nobody on the team can read SQL—you will rubber-stamp wrong joins. Skip if you need connected-account rollups on a platform. Skip if your urgent problem is write automation; this launch is analysis. And skip dumping finance data into a shared agent workspace with contractors until scopes and retention are set.

Governance that belongs next to the free window.

January 31, 2027 is not only a pricing date. It is a forcing function to decide which teams may burn Queries credits. Finance, growth and founders will all want "just one more prompt." Give each team a workspace, a scope and a monthly ceiling before February usage-based billing starts.

Keep a shared folder of blessed SQL. When an agent produces a useful reconcile query, save it beside the human-written Sigma note you already trust. Over time the folder becomes the real product—not the chat transcript.

If your Stripe account is a platform, say out loud that connected-account analytics are out of scope for this launch. Do not let account managers promise rollups the API will refuse.

Pair this note with your MCP key migration. Analytics read scopes on an agent key are a different safety story than a full-access secret that can also write. Do both tickets; do not merge them into one vague "Stripe AI" project.

Sequence of work

Read-only first, then the free window

  1. This week

    Connect MCP with analytics/report read scopes only.

  2. Next two weeks

    Shadow five recurring finance questions; save SQL.

  3. By Jan 31, 2027

    Decide Queries volume and budget after free window.

  4. Feb 1, 2027

    Usage-based Queries pricing begins; Reports stay free.

Dates from Stripe blog/docs Nov 5–6, 2026; rollout steps illustrative.

Put this into practice

Connect MCP with metrics and reports read scopes only. Run five prompts you already answer monthly. Save the SQL next to the answer. Keep write tools off. Calendar January 31 for the Queries free-window review. If you want help folding Stripe analytics agents into business automation without mixing keys and writes, get in touch with your current Stripe MCP setup and a sample monthly finance question list.

Keep a short evidence packet for each blessed query: the prompt, the SQL, the Stripe metric name if any, and who signed off. Board materials should cite that packet, not a chat screenshot.

When the free window ends, revisit which prompts still deserve live Queries credits versus a scheduled report. Not every executive curiosity needs an always-on agent.

If multiple brands share one Stripe account, label answers with the brand filter you used. Mixed-brand SQL is how wrong narratives travel.

Train one finance partner to read agent SQL even if they do not write it. The review skill matters more than the generation skill. A thirty-minute lunch-and-learn with three saved queries beats a generic "AI for finance" workshop.

Founders who want investor decks from agents should still pass the deck through the same human who owns the Stripe definitions. Speed is fine; unverified MRR is not.

Related work: Sutton Place.

twimco.

We build alongside the people doing the work.
Bring us the problem you’re thinking about.

Related work

Sutton Place ↗︎