Cube ☰
AI

Product

menu arrow

Core Platform

Platform Overview

Workflow-native FP&A, powered by AI

Cube Workspace

Your command center for financial intelligence

Data Engine

Powering structured, real-time planning

AI for FP&A

AI that does real work across your workflow

Extend & Connect

Integrations

Data you need, where you need it

MCP Server

Enterprise-grade, secure LLM integration

Solutions

menu arrow

Use Cases

Data Management

Reporting & Analysis

Planning & Modeling

Strategic Finance

Cash Management

Ad-Hoc Reporting

Annual Planning

Forecasting

Scenario Analysis

Variance Analysis

Roles

Cube for CFOs

Cube for FP&A

Cube for Controllers

Cube for CEOs

Industries

Business Services

Financial Services

Healthcare

Manufacturing

Nonprofits

Real Estate

Retail

SaaS / Software

Multi-Location

Resources

menu arrow

Get Started

Strategic Finance Assessment

Benchmark your finance maturity

Spreadsheet Templates

Ready-to-use models and tools

Developer Center

Build and extend with Cube’s API

Learn

Blog

Insights and tips for modern FP&A

The Finance Fix Newsletter

Weekly finance news and strategies

eBooks & Guides

Deep dives on strategic finance levels

FP&A Glossary

Key terms every finance pro needs

Webinars & Videos

Expert talks and product demos

Community and Support

Customer Stories

Real results from real teams

Help Center

Answers, docs, and how-tos

Workshops

Live sessions to level up

Company

menu arrow

About Cube

Explore our vision and team

Careers

Help shape the future of FP&A

In the News

Cube featured in top publications

Security

Enterprise-grade, cloud security

Partnership

Join the Cube ecosystem
Pricing
FREE DEMO
Cube MCP Server: The MCP Server for Finance | Cube
AI Assistants via MCP

The MCP server for finance: finance-grade answers in your AI assistant.

Cube is the MCP server for finance, connecting Claude, ChatGPT, Copilot, and any MCP-compatible AI assistant to governed, decision-ready data. Every answer traces to the source transaction.

Book a demo See it in action
An AI assistant conversation using the Cube integration: a full-year AI reforecast charted against budget month by month, finishing 2.46 million dollars ahead of budget, with the Cube connector and the cube-data-publishing skill in the context panel.

The problem

You live in your LLM. The numbers still need a login.

Your assistant already reaches the rest of the stack through MCP. Finance is the holdout, so the work stops while someone goes and fetches a number.

Everything else is connected

The CRM, the docs, the warehouse: all one prompt away through your other MCPs. The finance question is the one that breaks the flow.

The workaround is paste

So screenshots and CSV exports go into the chat, and the assistant answers from last Tuesday's numbers.

Pasted numbers are ungoverned

No permissions, no lineage. AI can be 80% right, and in finance that is 100% wrong.

Under the hood

How the Cube MCP Server answers.

Your assistant asks over an OAuth-secured connection. Charlie plans the work, routes it to the right agent team, and the answer comes back decision-ready from the Agentic Finance Layer, with the lineage attached.

Watch a governed answer land in the assistant.On the demo, we ask for March opex, drill to the Datadog true-up, and show the permissions doing their job.
Book a demo

Ask why

A variance deep-dive from a single question.

Ask why opex ran $160K over in March and the assistant walks the bridge. Cloud and software is the big step, $412K against a $328K budget, traced to the true-up behind it.

$2.63M +$35K +$84K +$41K $2.79M Budget Payroll & benefits Cloud & software T&E Actual opex · Mar 2026 · amber = over budget · axis starts at $2.50M · every bar opens to its transactions

Write-back

A reforecast, built in conversation.

An assistant grows the full-year forecast off actuals, charts it against budget, and publishes the scenario back to Cube, gated by the Use MCP (Write) permission. One prompt can also produce a complete board deck, and Cube Decks keeps those slides live. Behind every request, Charlie routes the work across the agent teams below.

Charlie plans and routes every request · the Cube MCP Server powers any agent you build

Why teams connect

Answers you can put in front of the board.

Does it solve my problem?

Real figures, no pasting

The assistant reads live from Cube. Nobody exports a CSV, and nobody answers from last month's snapshot.

How is it different?

Lineage travels with the answer

Every figure carries its context and traces to the GL transaction, so the follow-up question has somewhere to go.

Can I trust it?

Access follows the user

Role-based access is enforced on every query. The assistant can never show anyone a number Cube would withhold from them directly.

Bring your CFO and your IT lead.One demo answers both questions: what the assistant can do, and what it can never see.
Book a demo

Adoption

Setup your IT team will actually like.

The connection is OAuth-secured from Cube, with nothing to export and no developer resources required. It works alongside the AI tools your team already uses.

The Cube MCP Server deploys alongside your existing stack with OAuth-secured connections, SSO, and no code. Finance turns it on from Cube without an IT project.

Works with Claude logoClaude ChatGPT logoChatGPT Microsoft Copilot logoCopilot + any MCP-compatible assistant
2 min

OAuth setup

Connect an assistant in under two minutes.

100s

of sources

Hundreds of source systems feed the model, with pre-built connectors for major platforms.

0

Rip-and-replace

Cube deploys alongside your stack and the assistants you already run.

No code

Finance-led

Your team turns it on from Cube, and permissions stay in your admins' hands.

Get started

How to connect an AI assistant to Cube

1. Enable the MCP Server in Cube

Open the MCP Server panel in Cube settings and enable access for your workspace.

2. Authorize your assistant via OAuth

Authorize Claude, ChatGPT, or Copilot through the OAuth prompt. No keys, no exports.

3. Ask your first question

Answers come from your Cube model with permissions and lineage already enforced.

Two ways in

Use our agents, or build your own.

Use the FP&Agents

Charlie and the four teams ship ready to work. Ask from Cube, chat, or your spreadsheet through Cube's Patented Spreadsheet Connectivity, and the FP&Agents pick it up with governance already on.

Build on the Cube MCP Server

Your platform team can point any agent framework at the same governed model, with permissions and lineage on every query. Building on ungoverned data is a faster way to build the wrong thing.

Why Cube

The MCP server for finance, not another connector.

The best MCP server for finance is the one that understands finance. Unlike generic database MCP connectors, the Cube MCP Server serves answers from a trusted financial model, with role-based permissions, cell-level security, and transaction lineage enforced on every query. Generic connectors return rows; Cube returns decision-ready answers that trace to the source.

Generic MCP connectors

A warehouse or database MCP hands your assistant raw tables. It has no idea what "gross margin" means in your model, which version of the forecast is current, or who is allowed to see exec comp. The assistant answers confidently from whatever it can reach.

Building on raw APIs

Wiring an assistant to finance systems by hand means rebuilding permissions, metric definitions, and audit logging yourself, then maintaining all three. The Cube MCP Server ships that layer: OAuth setup in under two minutes, with your admins in control.

See the source systems that feed the model

Security & controls

Every answer passes governance before anyone sees it.

The CFO sees everything. A VP sees their cost center. An external auditor sees read-only with exec comp redacted. Same source, different access, enforced at the cell level.

SOC 2 Type II SSO & SAML Full audit trail Cell-level security

Role-based access on every query

The MCP connection inherits the user's Cube permissions. There is no separate AI access path.

"Use MCP (Write)" permission

Reading and writing are separate grants. Write-back requires its own dedicated permission from your admins.

OAuth-secured connections

Assistants connect through secure OAuth from Cube, with no keys to paste into chat tools.

Full audit trail

Every change is logged with who, when, and before and after values, and publishes can be undone.

SSO

Okta, Microsoft Entra, and Google, with SAML support.

Cell-level security

Sensitive lines like exec comp stay redacted for the roles you choose, down to the cell.

Read the full security overview

Trace to the source

Behind one answer.

The $412K your assistant just quoted opens to its accounts and the transactions underneath. The auditor's follow-up already has its answer.

every figure your assistant returns stands on this

See it on your own assistant.Claude, ChatGPT, or anything MCP-compatible. We connect it to a governed Cube model and you drill the answers.
Book a demo

Make every AI tool in the company finance-literate.

Thousands of finance professionals run on Cube. Book a demo and watch a governed answer land in your assistant, traced to the source.

Book a demo

Common Questions

Everything you need
to know about Cube's MCP Server

An MCP server for finance is a Model Context Protocol server that gives AI assistants governed access to a company's financial model, with permissions and transaction lineage enforced on every query. The Cube MCP Server is that layer for your Cube model: the assistant asks in plain language, and the answer comes from the same governed model Cube delivers everywhere work happens, with the source transaction one click away.
Claude, ChatGPT, Copilot, and any MCP-compatible AI assistant can connect. The connection is OAuth-secured and takes under two minutes to set up, with no exports and no developer resources required. If your team would rather ask in Slack or Teams, the same AI Analyst answers there.
Only with the write permission. Write-back is gated behind a dedicated permission called Use MCP (Write), granted by your admins. Everything else is read-only, and every change in Cube lands in the full audit trail with before and after values.
No. Access follows the user's permissions on every query, enforced down to the cell. An assistant acting for a VP sees that VP's cost center, and an assistant acting for the CFO sees everything, exactly as they would in Cube.
Pasted data is a stale copy with no permissions and no lineage. Through the Cube MCP Server, the assistant reads live from the governed model, your business logic and role-based access apply to every query, and every figure traces to the source transaction.
The Cube MCP Server is the part of Cube's APIs built for AI assistants. It uses the open Model Context Protocol, so any MCP-compatible tool can query the governed model without custom integration work.
Yes. The Cube MCP Server speaks the open Model Context Protocol, so your platform team can point any agent framework at the governed model. Custom agents inherit the same role-based access, audit trail, and transaction lineage the FP&Agents use, so every figure they return is one your finance team can stand behind.
The Cube MCP Server is generally available to all Cube customers, and usage limits may apply. Your admins control who can connect an assistant and who holds the Use MCP (Write) permission.
The best MCP server for finance is one that enforces permissions and lineage on every query rather than exposing raw tables.
 
The Cube MCP Server connects Claude, ChatGPT, Copilot, and any MCP-compatible assistant to a trusted financial model, so answers arrive decision-ready and trace to the source transaction.
 
Generic database MCP connectors return rows without context, permissions, or lineage.
logo-cube-image.svg logo-cube-image.svg

Schedule a Demo.

Please provide your details, and a member of our sales team will personally coordinate a time that works best for you.