Revenue context for AI agents · local-first

Your agent is confident.
Your CRM is wrong.

Point Claude straight at a raw CRM and it answers fast and wrong: duplicates double-count pipeline, won deals that never paid look like revenue, stale close dates poison the forecast. Unfurl syncs HubSpot and Stripe into a resolved revenue graph on your machine and serves it to any MCP agent, with caveats inside every answer. No AI credits: your own agent does the reasoning.

Get early accessTry the demo
Same data, same question

The difference is
a resolution layer.

claude + raw CRM APIwrong
> which customers are at risk?

You have 4 customers in HubSpot: Acme Robotics, Brightloop, Driftwave, and Glasswing Labs. None are flagged as at risk.

· Driftwave canceled in Stripe; the CRM never heard about it
· Glasswing's deal is closed-won, but it never started paying
claude + unfurlverified
> which customers are at risk?

2 of 4 customers need attention: Driftwave canceled (−$2,000/mo), and Glasswing Labs is closed-won in HubSpot with no active Stripe subscription.

Also: one Stripe customer ($900/mo, past due) matches no CRM account. Run data_health for details.

Sample data from unfurl demo. Same data, same question.

What it answers

Ask in plain language.
The agent picks the tool.

Real questions, and the read-only tool that answers each one.

Unfurl next to hosted platforms →

How it works

Sync. Resolve. Serve.

  1. SyncPull HubSpot companies, contacts, deals, and owners, plus Stripe customers and subscriptions, into a local SQLite file.
  2. ResolveDedupe companies by domain and contacts by email, link billing to CRM, and log whatever cannot be fixed.
  3. ServeSix MCP tools over stdio. Your agent gets resolved numbers with caveats, not raw records to misread.
  4. AskIn Claude or any MCP client: "which won deals never started paying us?"
Six tools

Each one with its
caveats built in

Your agent picks the tool. Answers say what was merged, what is missing, and how much to trust the number.

pipeline_summary
Open pipeline by stage or owner, from the resolved graph. Duplicates are merged and deals with missing amounts are flagged, so totals are not silently low.
win_loss
Win/loss for deals closed in a period: counts, values, win rate, average cycle length, and recorded loss reasons.
account_lookup
The single resolved view of one account: people, deals, subscriptions, and active MRR, with CRM duplicates merged and billing linked.
churn_signals
Churn and revenue risk across CRM and billing: canceled or unpaid subscriptions, past-due accounts, and deals marked won that never started paying.
revenue_summary
Current MRR and ARR from billing, by subscription status, with revenue that could not be attributed to a CRM account called out.
data_health
The trust report: duplicates merged, orphaned records, missing amounts, stale close dates, and when each source last synced.

All tools →

Use cases

RevOps questions,
answered from the truth

Fix what the CRM never told you
Pipeline hygiene
Duplicates merged, orphan deals and missing amounts flagged, before anyone quotes a number.
data_healthpipeline_summary
Clean inputs, honest forecasts
Forecast inputs
Stale close dates and undercounted deals surfaced, so the forecast starts from clean inputs.
pipeline_summarydata_health
Why you win, why you lose
Deal reviews
Win rate, cycle length, and loss reasons, plus the full resolved account behind any deal.
win_lossaccount_lookup
Won is not paid
Sales-to-billing handoffs
Deals marked won in the CRM with no active subscription in billing, found automatically.
account_lookupchurn_signals
See risk before it leaves
Renewals and churn
Cancellations, unpaid and past-due subscriptions, and MRR by status, straight from billing.
churn_signalsrevenue_summary
Built to be trusted

Four things
you can check yourself

  • Local-first.Everything lives in a SQLite file on your machine. Nothing is sent to us.
  • Read-only credentials.A HubSpot private-app token and an optional restricted Stripe key. Unfurl never writes to your CRM.
  • Agent-neutral.Anything that speaks MCP gets the same resolved graph: Claude, Cursor, or whatever comes next.
  • Caveats in the answer.What could not be fixed travels with the number, not in a dashboard nobody opens.
For small revenue teams

Too small for a RevOps hire.
Too busy for a sales call.

Install in minutes. Try it on sample data before you connect anything.

Get early accessRead the quickstart