Introducing the Lili MCP server — and why we built it for accountants first.
Every accountant with a real client book knows the drill. You’re constantly switching between portals — logging in, pulling statements, exporting transactions, missing transcations, emailing clients, waiting. Not once a month. Every time a question comes up. Every time a close approaches. Every time a client calls.
If you have 20 clients, that’s 20 portals, 20 login sequences, 20 export flows — repeated across every cycle, every question, every request. The average accountant burns over 30% of their time on manual, repetitive administrative work instead of time spent advising.
Lili already brings much of that work into one place: transactions, reports, invoices, bills, tax tools, and real-time financial visibility. That foundation makes the work more organized and gives accountants a clearer view of each client’s business.
But even with a clearer view, accountants still had to know where to look. They had to go into the platform, check each client, identify what mattered, and pull the answer together themselves.
So we went a layer deeper. What if instead of navigating to the data, your AI swept your entire book every morning and surfaced exactly who needs attention — the client with a Q2 tax shortfall building quietly, the one with a $2,200 duplicate charge they haven’t noticed, the one whose cash runs tight in six weeks if one invoice comes in late? What if the flag, the analysis, and the drafted client email were all waiting for you before anyone called?
That’s what we built. A real-time intelligence layer across your entire client book, delivered through whatever AI you’re already working in. You show up knowing things your clients don’t know yet. You catch problems before they become calls. Your AI navigates. You advise.
For the first time, you can have visibility across every client — all the time, not just when you have a moment to log in.
Why this is possible now
For years, connecting an AI tool to a banking platform meant a custom integration — one AI client, one bank, one bespoke API contract. It wasn’t scalable and it wasn’t open. The accountant was still stuck with whatever integrations their bank had chosen to build.
That changed with the Model Context Protocol (MCP). MCP is an open standard — think of it as a universal connector between AI assistants and external data sources. Any AI assistant built to the standard can talk to any MCP server. Build the server once, and every compatible AI client works automatically.
That’s what made this moment the right moment. We built one server, and opened it to everything. The accountant uses whatever AI they already trust.
What we built
Today we’re launching the Lili MCP server.
Once an accountant connects the Lili MCP to their AI assistant, they can ask questions in plain language and get real answers — backed by live banking data. Starting with the question that matters most:
→ What needs my attention across my portfolio this morning?
That one prompt sweeps an accountant’s entire client book and returns a prioritized list — who has uncategorized transactions, who is under-reserved for their next estimated tax payment, whose cash balance is running thin against upcoming obligations. What used to take two hours of portal-hopping now takes seconds.
From there, the AI goes as deep as needed:
→ Run me Rivera Creative’s Q2 tax position. What do they owe, what have they saved, and what’s the gap?
→ What’s their cash position over the next 90 days if two of their three open invoices come in late?
→ Draft an email to the client explaining the shortfall and recommending a transfer to their Tax Bucket.
→ Show me the uncategorized transactions from May and flag anything that looks like a duplicate.
The AI does the navigation. The accountant does the advising.
What it looks like in practice
A solo bookkeeper managing 28 clients used to lose the better part of two hours every morning just switching between portals — logging in, pulling transactions, flagging what needed attention, starting over. That time is now one prompt. But the bigger change isn’t the two hours. It’s what the AI surfaces that the bookkeeper wouldn’t have found by portal-hopping at all: the client whose spending spiked 40% last week with no corresponding revenue, flagged before anyone asked. The bookkeeper shows up to every client conversation already knowing what matters.
A small-firm CPA doing month-end close for a marketing agency used to spend the first hour just assembling the picture — pulling the P&L, checking the tax bucket, reviewing open invoices. With the MCP, that assembly takes ten minutes. So now, what happens in the time that’s freed up? The AI has already modeled two cash flow scenarios, calculated the Q2 tax gap, and drafted the client email with the numbers. The CPA reviews it, adjusts the tone, and sends it. The client gets a proactive call about a shortfall they didn’t know was coming. That’s not a faster accountant. That’s a more strategic advisor.
A fractional CFO is on a call with a client when the client mentions a charge that doesn’t look right. Two identical payments, same vendor, same amount, posted eighteen minutes apart. In the old world, that’s a “let me look into it and get back to you” moment — which means a follow-up email, a portal login, a screenshot, and a delay. With the MCP, the CFO asks mid-call and has the answer in seconds: confirmed duplicate, $2,200, we’re disputing it. The client didn’t even know it had happened. That’s not faster service. That’s a different category of service entirely.
A tax advisor managing a high-stakes portfolio of C-Corps used to break the momentum of every planning call to look something up — logging in separately, finding the number, getting back on track. Those interruptions are gone. But more importantly, the advisor now arrives at every call with a pre-built picture: current balances, recent payroll, any anomalies in the last 30 days. The conversation starts at a higher level because the data work was already done. Clients notice. They feel like they have an advisor who actually knows their business.
The AI does the navigation and the pattern recognition. The accountant does the advising.
On security
The MCP server is a thin, secure adapter between the AI client and Lili’s existing banking infrastructure. It uses the same internal APIs and the same authorization model we’ve always used — identity verified via OAuth 2.0, access tokens scoped to the authenticated user. Accountants can only access client accounts where the client has explicitly granted them permission, the same as the Accountants Portal today.
No new surfaces. No new permissions. The AI assistant is a new way to talk to data that was already theirs.
What’s next
The Lili MCP server is live today for all accountant partners. It works with any AI assistant that supports MCP.
We’re continuing to expand the tool set based on what accountants actually ask for. If you’re an accountant partner and there’s a workflow you want to automate, we want to hear about it.
The accountant’s job is to advise — to see around corners for small businesses, to catch what’s coming before it arrives, to be the person in the room who actually understands the numbers. That job has always been limited by how much time the data work left over. This is our attempt to change that ratio — and to give accountants a level of visibility across their book that wasn’t possible before.
