- Announcement
TL;DR
- Primary is a treasury layer that sits on top of your finance stack. It is a platform that consolidates all of your accounts for real time cash visibility, multi currency management, investment of idle cash at best in market rates and FX strategy and hedging.
- Primary's MCP server is a layer that connects your company’s live treasury position with your LLM of choice.
- Connect your back end finance stack once with Primary and your finance team can ask the AI tool they already use questions like ‘what balance of USD do we have across all accounts and do we have enough to cover payroll on the 15th?’ or ‘what’s our FX exposure in the next 6 months?’ and receive live answers instead of last Tuesday's CSV.
- Then the next layer is execution. You can ask your LLM to prepare payments, queue conversions and set up hedges, all routed through your approval queue inside Primary.
- This is the new finance stack.
What is MCP?
MCP is a protocol that lets an LLM securely connect to third-party platforms. When you connect Primary's MCP, the LLM can read your treasury data in real time to answer questions, surface insights, and help you make faster decisions.
Why we built it
A finance team at a growth-stage company spends most of its week answering questions that should be instant. The information on common queries like ‘can we cover payroll’ or ‘what is our FX exposure next month’ is all available but it sits across multiple tools or accounts.
The workflow underneath those questions is six bank portals, a manual reconciliation across entities, and a check to make sure the data is correct.
AI can easily solve this problem as long as it has the right context. The problem right now is that it can’t see your accounts, your entity structure, or the currencies you're holding and even if you provide this data, it is not updating in real time.
Primary closes that gap by consolidating all of your finance data and the MCP allows you to find answers fast.
We built this because it is the next iteration of the finance stack.
Primary sits on top of your existing finance stack, connecting all of your accounts so data is accessible in real time. The MCP connects to your LLM of choice so you can ask query questions relating to treasury with natural language. Then you can execute in platform.
How the MCP works
Three things are happening in the background every time your AI tool answers a treasury question:
Current data goes in
Primary maintains a live connection to your bank accounts and treasury counterparties. When the AI asks for a balance or an exposure, it's pulling from positions that are accurate to the minute rather than a daily export.
Best-in-class intelligence is applied
Primary layers your treasury policy on top of the raw data. The hedging framework you defined, the approval thresholds you set, the cash-sweep logic you've codified already in platform. All of that travels with the data into the AI's response.
Ready-for-approval actions come out
When the conversation moves from analysis to execution, the AI prepares the instruments (i.e. forwards, options, payments, conversions) and queues them inside Primary for human approval. Nothing executes until a person with the right permission signs off.
A use case: FX exposure and hedging strategy
You open your tool of choice and ask:
What's our FX exposure in AUD over the next 12 months? And what hedging strategy would you recommend?
With the Primary MCP connected, the AI returns a live answer grounded in your company's actual position:
Your net AUD exposure over the next 12 months is approximately -$4.2M (you're a net payer in AUD terms, primarily driven by your Sydney office costs and two supplier contracts).
Here's how it breaks down:
- USD receivables converting to AUD: +$8.1M
- AUD-denominated costs (payroll, rent, local suppliers): $12.3M
- Net short AUD position: $4.2M
What I'd recommend:
Hedge 70% of the exposure (~$2.9M) and leave the rest open. Here's the split:
- Forward contracts for the payroll and rent obligations: these are certain, so lock them in.
- FX options for the supplier contracts: there's some volume uncertainty there, so options give you the floor without locking you into notional you might not need.
- Leave the remaining 30% unhedged as a natural buffer against your USD receivables moving.
A few things to notice about that answer. It's not generic advice. The numbers are the company's actual numbers. The reasoning - hedge the certain liabilities with forwards, the uncertain ones with options, keep a natural buffer - is the treasury policy the finance team had already codified inside Primary.
The execution layer
The second part to this is execution, with human controls.
The Primary MCP separates preparation from execution. The AI can prepare actions; only an authorised human can release them.
Here's how it looks in practice, following on from the FX conversation above:
CFO: Set up in Primary all the hedges that we'll need over the next 12 months.
AI: On it. Based on your exposure, I'm booking:
$2.9M across 10 instruments via ANZ Markets. Margin draw: ~$145k.
Log in to Primary to confirm.
Nothing moves until the controller in Primary approves the queue.
- 6 forwards: $2.1M - payroll & rent (Mar–Aug)
- 4 FX options: $0.8M - supplier contracts (Jun–Mar)
What this changes for the finance team
The job moves up the stack. There is less time looking for answers and more time for setting policy, asking the sharper question, and focusing on growth.
The work that disappears is the assembly work: copy-pasting balances, building exposure models from scratch, chasing payment confirmations across banks. The work that remains is the work that actually requires judgement.
FAQs
1. Which AI tools does the Primary MCP work with?
At launch: ChatGPT, Claude and Gemini. We're adding support for additional MCP-compatible tools as the ecosystem matures.
2. Is the MCP really free? What's the catch?
Primary’s introductory tier is free. This provides read only MCP that connects all your accounts and allows for live data to be connected to your AI tool of choice.
3. What are the pricing tiers?
- Free tier: read-only MCP. Connect all accounts + balances into Primary and use LLM to query.
- $399/mo: execution capability.
- Enterprise: for customised experience and access to higher tier investment accounts.
4. Is my data secure?
Yes. Claude accesses your data under your existing Primary permissions. Data is not stored by Claude between conversations.
5. What can the AI execute on my behalf?
The AI can prepare payments, currency conversions, hedges (forwards and options), and cash sweeps. It cannot execute any of those without a human approval inside Primary. The approval rules are whatever you've configured in platform.
7. What if the AI gives a wrong answer?
The architecture is designed to make this difficult. The data the AI works from is live and sourced directly from your banking partners. The reasoning is also constrained by your treasury policy (i.e. rules you have set up in Primary’s platform). And no action executes without a human approval.
That said, you should treat the AI's analysis the same way you'd treat any analyst's first draft: useful, fast, and worth a sanity check before you act on it.
8. How is this different from a Treasury Management System?
A traditional TMS is a piece of software you open to do the work. The MCP makes Primary a layer that operates underneath whatever AI tool your team already uses. You don't replace your workflow with Primary; you give your workflow live access to your treasury.
9. Can we customise the treasury policy and hedging framework?
Yes, through Primary’s platform. Your finance team encodes the things it already knows intuitively, i.e. FX thresholds, hedging philosophy, approval chains, and the MCP makes that policy travel with the data.
