A secure connection to your data

Give Claude your data, and ask it anything

A secure connection between your data and the Claude tools your team already has open. Ask in plain words and get a working answer back, with the method shown.

Let’s chat See the engagement

What it looks like

What comes back when you ask

A working report, not a paragraph: figures, filters, an export, and a note on how it was built. Four examples, all sample data.

Asked: Which customers have not ordered in one to five years?

Customers who have not ordered in 1 to 5 years

window: ≥365 and <1825 days inactive · cutoff 2026-06-02 · qualifying states sale, done · rolled up to parent account · deduped by name · 40 confirmed-closed accounts excluded

Filters work here. Showing the 280 most recent of 4,936 inactive accounts, deduped, with 40 confirmed-closed accounts removed so reps do not chase businesses that no longer trade. For the full list use the live view.

View full live data →

4,936

Inactive total

571

Wholesale

2,261

Retail

1,703

No rep assigned

2016

Oldest account since

Search company, contact, city…All channels ▾All reps ▾All states ▾Export CSV

280 matching · sample of 280 · 4,936 in total

By channel

Public225

Wholesale37

Retail18

Top regions

Northeast30

West23

Midwest21

Southeast17

Illustration of the shape of a real report. Every figure and label here is invented sample data.

Asked: Which product lines are selling below cost?

Product lines ranked by contribution margin

window: rolling 90 days to 2026-06-02 · landed cost includes freight and duty · customer rebates accrued to the line that earned them · intercompany transfers excluded · 3 lines with no cost record shown as unqualified

Two lines are under water. Ranked on contribution after freight, duty and accrued rebates. Averaging these into a category figure hides both of them, so they are called out instead.

Open the live view →

Lines below cost

18.4%

Blended margin

31.4%

Best line

Cost record missing

Search company, contact, city…All categories ▾Last 90 days ▾Export CSV

412 lines evaluated · 26 below cost · 3 unqualified

Lowest contribution, after freight and rebates

Line	Revenue	Landed cost	Margin	Status
Bordeaux 750ml, flint	184,200	189,000	-2.6%	Below cost
Crown cap, gold 26mm	96,400	85,800	11.0%	Thin
Cork, natural 44 x 24	142,700	115,300	19.2%	Healthy
Champagne 750ml	318,900	233,400	26.8%	Healthy
Claret 750ml, amber	271,500	186,200	31.4%	Healthy

Illustration of the shape of a real report. Every figure and label here is invented sample data.

Asked: Who owes us money, and how late is it really?

Receivables by age, with the buckets labeled honestly

as at 2026-06-02 · aged from invoice due date, not issue date · credit notes netted against the invoice they relate to · disputed balances flagged, not removed · 2 accounts in payment plans shown separately

Not-yet-due is its own column. Most aging reports fold not-yet-due in with recently-late, which makes the current bucket look healthier than it is. Here the split is explicit.

Open the live view →

Days sales outstanding

116.6k

Past due

Accounts over 60 days

On payment plans

Search company, contact, city…All divisions ▾Past due only ▾Export CSV

184 open accounts · 31 past due · 2 in dispute

Largest past-due balances

Account	Balance	Past due	Days	Status
Meridian Hospitality Group	84,120	61,400	74	Chase
Cobalt Retail Partners	52,940	38,210	61	Chase
Harbor Line Distributors	41,300	12,880	38	Watch
Fairmont Beverage Co	29,760	4,120	19	Watch
Selwyn Hotels	18,450	0	8	Current

Balance by age

Not yet due214

1 to 3063

31 to 6034

61 to 9019

Over 9012

Illustration of the shape of a real report. Every figure and label here is invented sample data.

Asked: Which suppliers are actually delivering on time?

Supplier on-time and in-full, accumulated over 12 months

window: 12 months to 2026-06-02 · on-time measured against the confirmed date, not the original request · partial deliveries counted as late until complete · 1 supplier with fewer than 5 orders excluded as not comparable

Measured against the confirmed date. Scoring against the date you originally asked for flatters nobody and blames everybody. This accumulates every order instead of being assembled before a review.

Open the live view →

86%

On time, in full

31 d

Longest lead time

Suppliers below 80%

Too few orders to score

Search company, contact, city…All categories ▾Last 12 months ▾Export CSV

17 suppliers scored · 1 excluded · 2,140 orders

On-time and in-full by supplier

Supplier	Orders	On time	In full	Status
Ashgrove Packaging	186	61%	74%	Review
Harbor Line Glass	243	78%	88%	Watch
Cedar Ridge Closures	412	91%	94%	Good
Northvale Glass	588	96%	97%	Good

Average lead time, days

Northvale12

Cedar Ridge19

Harbor Line24

Ashgrove31

Illustration of the shape of a real report. Every figure and label here is invented sample data.

1 of 4

How it is put together

01
It speaks MCP
Model Context Protocol is Anthropic’s open standard for connecting AI to real systems. Claude Connect is an MCP server over your data, so anything that speaks MCP can reach it.
02
Across the whole Claude ecosystem
Desktop, Code, claude.ai and Slack all reach the same data. Your team keeps whatever they already use.
03
You decide what it can reach
It exposes the slice you grant it and nothing else. Read-only by default, against the tidied copy.

Who Claude Connect is for: anyone on your team with a question about the business

An MCP server over your own data

Works across the Claude ecosystem

You grant the scope

Answers state their own method

What comes back

Reports that tell you how they were built

Ask a broad question and you get a working report, not a paragraph: figures, charts, a filterable table, an export. Each states its window and cutoffs at the top.

The method is on the page
Which tables it read, the date range, the rules applied, the records left out. Check it instead of trusting it.
Filters that work
Search, dropdowns, date ranges, export. A working view of your data, not a screenshot of one moment.
It tells you what it left out
Where a list is capped, or records were excluded deliberately, the report says which ones and why.
Built by people who do this
We run MCP servers in production against real customer systems.

Why it holds up in production

It works with Radiance

Radiance watches the numbers on a schedule. This lets anyone dig into the same data, on the same definitions, whenever they need to.

It works from your data

Answers get built from your records and show what they drew on. Where the data cannot support a claim, it says so.

No new tool to adopt

Nobody learns an interface. The assistant your team already uses just knows the business now.

See the whole engagement

The other half

Both products, one data foundation

Radiance

A screen per question, search that reaches the paperwork behind them, and the routine checks running overnight. Whatever you look at every week, already up when you open it.

Discover more →

Start with the numbers you cannot see today

Half an hour works out which measures are worth baselining and whether your systems can be read.

Let’s chat

Give Claude your data, and ask it anything

What comes back when you ask

Customers who have not ordered in 1 to 5 years

Product lines ranked by contribution margin

Receivables by age, with the buckets labeled honestly

Supplier on-time and in-full, accumulated over 12 months

How it is put together

It speaks MCP

Across the whole Claude ecosystem

You decide what it can reach

Who Claude Connect is for: anyone on your team with a question about the business

Reports that tell you how they were built

The method is on the page

Filters that work

It tells you what it left out

Built by people who do this

Why it holds up in production

It works with Radiance

It works from your data

No new tool to adopt

Both products, one data foundation

Radiance

Start with the numbers you cannot see today