The Evolution of AI for Business, and Recursyv’s Part in It

How MCP connectors are turning Claude into your business’s best-informed colleague

Three years ago, most of us met AI as a clever chatbot. It could write a poem, summarise a news story or explain a tax rule, but ask it a question about your business and it had nothing to say. It had been trained on the public internet and a library of documents that had never heard of your clients, your service desk or your sales pipeline. It couldn’t even check today’s news. Claude only gained access to the live internet in March 2025, at first just for paying US subscribers, and it wasn’t until late May 2025 that every user could use it. That was less than 18 months ago.

That has changed. AI can now connect, securely, to the systems your business actually runs on: your service desk, your CRM, your knowledge base, your email and your calendar. It can answer questions that span all of them at once. In this article we look at how we got here, what it means in practice for businesses of every size, and what to watch out for as you make the move.

From chatbot to everyday habit

ChatGPT launched at the end of November 2022 and, according to analysts at UBS, reached an estimated 100 million users in about two months, then the fastest growth of any consumer app on record. Claude, Gemini and Copilot followed. By late September 2026, ChatGPT alone was reported to have around 1.2 billion weekly users.

For consumers the habit has shifted fast. Instead of trawling ten blue links, people ask an assistant and get an answer. In 2024 Gartner predicted that traditional search engine volume would fall 25% by 2026 as people turned to AI chatbots and virtual agents. Shopping is moving the same way: Adobe recorded a 693% year-on-year increase in traffic from generative AI tools to US retail sites over the 2025 holiday season. Research, comparison and product discovery that used to take an evening of browser tabs now take one well-phrased question.

But at work, something was missing. The assistant was brilliant at general knowledge and blind to the information that matters most to a business. To get value from it, people copied and pasted: an export from the service desk here, a spreadsheet from the CRM there, an email thread pasted in at the end. Every answer was only as good, and as current, as whatever someone remembered to paste.

The missing piece: your own data

In November 2024 Anthropic, the company behind Claude, introduced the Model Context Protocol (MCP), an open standard for connecting AI assistants to business systems. Anthropic describes it as a kind of USB-C port for AI: one standard way to plug an assistant into the tools a business uses.

We won’t dwell on the technology. What matters is that it caught on. Within a year MCP had been adopted by ChatGPT, Gemini, Microsoft Copilot and others, more than 10,000 public MCP servers were in use, and in December 2025 Anthropic donated MCP to a new Agentic AI Foundation under the Linux Foundation, so that no single vendor owns it.

The result is that the AI your team already uses can now look things up in your systems, live, and act on your behalf, within the permissions you set. It’s the same shift consumers have already made, now happening inside the business.

You don’t need a transformation budget

When a technology this significant arrives, the assumption is usually that only large enterprises can afford to use it well: the ones with data teams, BI platforms and seven-figure change programmes.

This time it’s different. The AI assistant is already a monthly subscription. The connectors are now a managed service too. With Recursyv AI, connecting your team’s AI to your business systems starts at £99 / $139 per user per month. Smaller businesses are moving quickly: according to the British Chambers of Commerce, 54% of UK SMEs were actively adopting AI in 2026, up from 35% a year earlier.

Our first two published case studies make the point. Both are US managed service providers that wanted results without an enterprise-scale project. Pacific IT Support, an MSP with about half its team offshore, recovered more than 200 hours in about two months. e-Coast Systems, a managed security service provider (MSSP), is saving more than 100 hours every month on billing and data audits. Both simply connected Claude to their Autotask service desk through Recursyv AI (more on both below).

What this looks like with Claude, day to day

At Recursyv we’ve chosen Claude as our own AI platform, so that’s what we’ll use for the examples here. Other AI providers offer similar capabilities, and Recursyv AI works equally with Microsoft Copilot, ChatGPT and Gemini, so you can choose the assistant that suits your business.

Once Claude is connected to your service desk (ITSM or PSA), your CRM, your knowledge base and Microsoft 365, you simply ask. Here are a few of the things our customers and our own team do every day.

Plan my day. A daily itinerary that understands your business, not just your calendar:

Illustrative example: in less than a minute, Claude comes back with something like this. “09:30 – Kestrel Logistics quarterly review. Four open tickets, one P2 waiting on a part since Tuesday. Renewal due in 62 days, with an £18k opportunity at proposal stage. Their finance director emailed on Monday about a laptop refresh; there’s no ticket for it yet.”

Answer the questions nobody has time to run a report for.

Audit, and protect every billable pound. This is where e-Coast Systems has found its biggest wins. A single prompt finds revenue that would otherwise slip through:

You can even upload a screenshot of a supplier invoice or a vendor’s licensing dashboard and ask Claude to check it against your service desk data, client by client, catching licence counts that no longer match what you bill.

Draft reports and briefings before anyone asks for them. Weekly service reports, quarterly business review packs, an account summary before a renewal call, a summary of every client email that describes a problem but has no ticket. Claude pulls the live figures, writes the narrative and produces the document, deck or spreadsheet. Your team reviews it rather than building it.

Take action, safely. With the right permissions in place, Claude can update records too: reassigning tickets when someone is off sick, logging notes, merging duplicate alerts onto the right client, flagging stale tickets. Every change is attributed to the person who asked for it.

None of this needs your team to learn a new tool, a report builder or a query language. If they can describe what they want, they can get it.

New in Claude: scheduled tasks and live dashboards

Two recent Claude features turn “ask when you need it” into “already done when you arrive”.

Scheduled tasks

Scheduled tasks let Claude run a piece of work automatically at set times. Anthropic introduced them in February 2026, and they now run remotely, so a task runs on schedule even when your computer is asleep or the app is closed.

How to use them: in a Claude conversation, describe the task and how often it should run, answer any questions Claude asks, then click Schedule. Tasks can run hourly, daily, on weekdays, weekly or on demand, and each one uses the connectors you’ve already set up. You can see past and upcoming runs, pause a task, edit it or run it now. Scheduled tasks are available on Claude’s Pro, Max, Team and Enterprise plans, with admins controlling access on Team and Enterprise.

A few schedules we’d suggest to any service business:

• Weekdays, 07:45. Morning brief: today’s meetings with account briefings, overnight P1s, tickets close to SLA breach, and client emails awaiting a reply.

• Every hour. SLA watch: tickets within two hours of breach, and stale tickets that haven’t been touched in three days.

• Mondays, 08:00. Service summary: last week’s volumes, resolution times and top clients by ticket count, ready for the team huddle.

• Fridays. Account health digest: for each account manager, clients with rising ticket volume, overdue invoices or renewals inside 90 days.

• Last working day of the month. Billing and margin review: unbilled and unapproved time, deliveries without a shipping charge, and work outside contract scope.

The benefit: the reports that used to take someone half a day, or never got done at all, are waiting for you when you sit down. Pacific IT Support uses exactly this kind of routine to check SLA status throughout the day and prepare weekly scorecards, without anyone monitoring the queues by hand.

Dashboards

On 8 October 2026 Anthropic launched Claude Dashboards in beta. You ask a question in plain English and Claude builds a dashboard that updates as your data changes.

How to use them: make sure your connectors are switched on, then ask, for example:

You only build it once. Every dashboard is saved automatically under Artifacts in Claude’s left-hand sidebar, so next time you simply open it rather than asking again, and Claude keeps it current as your data changes. Each chart shows when its data was last refreshed. Click any number to see exactly how it was calculated, and ask Claude to explain it.

Share it with the team. Dashboards start private. Click Share to give named colleagues, or everyone in your organisation, view, comment or edit access, and they open it from the link. Dashboards are in beta on Claude’s Pro, Max, Team and Enterprise plans; on Enterprise, an admin needs to turn them on.

The benefit: the live operational view that used to need a BI project, a data warehouse and a consultant now takes one sentence. The monthly PDF report that was out of date the day it was sent becomes a saved, shared, always-current page that the whole team can just open.

The compounding effect: more visibility, more value

A single connector already pays its way. Both of our case-study customers achieved their results with just one system, Autotask, connected to Claude. But every system you add widens what your AI can see, and the value grows faster than the number of connections, because it changes the kinds of question you can ask.

Research published in Harvard Business Review found the average worker toggles between apps and websites nearly 1,200 times a day, losing just under four hours a week (about 9% of their working time) to reorienting after each switch. Microsoft’s 2025 Work Trend Index found workers are interrupted up to 275 times a day, and that 80% say they lack the time or energy to do their job well. Each system on its own only adds to that.

We sum it up like this: one system answers, two systems correlate, three systems decide.

• One system (say, Autotask) answers questions about tickets, time and contracts.

• Two systems let you correlate. Service desk plus CRM: “Which accounts have open opportunities but falling satisfaction?” Service desk plus RMM: “Do repeat tickets at this client trace back to ageing devices?”

• Three systems let you decide. “Which clients have more than 10% of devices past end of life, rising ticket volumes and a renewal inside 90 days?” That’s a hardware refresh conversation, timed for the moment the contract is on the table.

Add Microsoft 365 and the AI gains the one thing your systems of record never hold: what the client actually said. “Which client emails this week describe a problem with no ticket?” Add your knowledge base (IT Glue, Confluence or SharePoint) and it knows how you fixed the problem last time.

The maths works in your favour. Three connected systems don’t give you three sets of answers. They give you seven possible combinations, each answering questions none of the systems could answer alone. Add a fourth and it’s fifteen. That’s why Recursyv AI includes up to three connected apps per user in the starting price: the value is in the combinations.

Pacific IT Support has already said it can add its RMM and CRM tools later. That’s where the compounding really starts.

The results so far

Pacific IT Support: 200+ hours recovered in about two months. Pacific IT connected Claude to Autotask through Recursyv AI. Claude now helps keep queues clean (spotting spam and duplicate tickets and consolidating recurring alerts), checks SLA status through the day, flags stale tickets and work ready to close or bill, reviews time-entry accuracy, dispatches recurring maintenance tickets and produces weekly scorecards.

“We are recovering at least 200 hours back for a small team in about two months. Given half our crew is offshore and dispersed, this has been a real force multiplier keeping our queues clean without someone monitoring them manually all day.”

— Erik McFrazier, President, Pacific IT Support

e-Coast Systems: 100+ hours saved every month. For an MSSP juggling security licensing, hardware deliveries, monitoring alerts and hundreds of client records, small gaps add up fast: a missed shipping charge, a licence that no longer matches the invoice, a device sitting on the wrong client. Auditing all that used to mean hours of cross-checking between Autotask, invoices and vendor dashboards. With Claude connected to Autotask through Recursyv AI, e-Coast now runs those audits with a handful of prompts. It checks licence counts against invoices and vendor dashboards, finds unbilled deliveries in seconds, merges monitoring alerts onto the right client, confirms configuration items are active and on the right client, cleans up duplicate contacts, and even suggests stock levels and what to market next quarter from real sales history.

“Recently I have been using it to do many different types of audits, from configuration items being active and on the right client, to billing audits for licensing. Honestly the time savings are tremendous!”

— Ryan Fogarty, Director of Operations, e-Coast Systems

Between them, these two service providers are each getting back around 100 hours a month: more than half a full-time member of staff each, put back into serving clients, and in e-Coast’s case revenue that would otherwise have leaked away.

These results are in line with the wider evidence. Anthropic’s own analysis of real Claude conversations estimated that AI cut the time for the tasks studied by around 80%, on tasks that would otherwise take about 90 minutes. Anthropic is clear that this is an estimate with caveats: it doesn’t count time spent checking the output, and controlled trials have found smaller gains. Even so, the direction is clear.

A year from now

Gartner predicts that 40% of enterprise applications will include task-specific AI agents by the end of 2026, up from under 5% in 2025. By 2028 it expects a third of user experiences to move from individual applications to AI-driven front ends, with agents working across applications so people no longer have to use each one separately.

So what does October 2027 look like for a business that gets this right?

• Nobody starts the day by opening five systems. The morning brief is already there: meetings, risks, priorities and the emails that need a reply.

• Monthly reporting is gone. Managers open a live dashboard, and the narrative for the board or the client review is drafted for them.

• Account managers walk into every meeting fully briefed, and spot the renewal risk or refresh opportunity weeks earlier than before.

• The service desk runs ahead of the SLA rather than behind it, with stale tickets, duplicates and billing gaps caught every hour, not at month-end.

• Every licence, delivery and site visit is billed, because the audit runs itself.

• New starters are productive in days, because they can ask the business questions in plain English rather than learning five interfaces.

• The business grows without the same growth in headcount, with people’s time going to clients and complex work, not chasing data.

The gap between businesses that do this well and those that don’t will be visible within a year, in response times, in renewal rates and in margin.

Why your choice of provider matters

All of this depends on giving an AI assistant access to your most sensitive business data. That makes the connector the most important security decision in the chain.

It’s tempting to download a free MCP server from GitHub or a public package library and plug it in. The evidence says to be careful:

• Weak authentication is the norm. An October 2025 review of more than 5,200 open-source MCP servers by Astrix Security found 53% relied on long-lived static API keys or tokens, and only about 8.5% used modern OAuth sign-in.

• Malicious copies are real. In September 2025 researchers found a fake version of a popular email MCP server, downloaded around 1,500 times a week, that quietly copied every email it sent to an attacker.

• Even big names get it wrong. A flaw in one major software vendor’s MCP integration could have exposed one organisation’s data to another, and the feature was taken offline for nearly two weeks. In another case, researchers showed that hidden instructions planted in a public GitHub issue could trick an AI agent into leaking data from private repositories.

An MCP server built as a weekend project by an individual developer is rarely designed around tenant isolation, per-user permissions, audit trails or the API limits of the system it connects to. It also comes with no support when a vendor changes its API.

Recursyv’s part: a decade of connecting critical systems

This is where Recursyv’s history matters. For the last ten years we have been integrating business platforms for MSPs and their enterprise clients, using our own proprietary integration platform hosted on Microsoft Azure. Hundreds of clients, from MSPs to banks, stock exchanges, healthcare providers, retailers and global manufacturers, rely on it to connect their critical service desk environments across organisational boundaries. When something breaks, the ticket, every update and the resolution flow between both teams straight away, so issues are resolved fast, whichever side of the relationship they start on. The platform has delivered 99.99999% uptime across ten years in production.

When MCP arrived, we saw that the same expertise could connect those systems to AI. In 2025 we launched Recursyv AI. It runs on the same Azure infrastructure, is built by the same engineers and is held to the same standard:

• Certified and accountable. ISO 27001 and Cyber Essentials certified, hosted on Microsoft Azure, and built to align with Microsoft’s Copilot security and governance standard, whichever AI you use.

• Zero data at rest. Your business data passes through Recursyv AI; it isn’t stored there. Each customer is fully isolated, and everything is encrypted in transit.

• Per-user permissions. You decide whether the AI acts with each user’s own rights in the connected app, with a least-privilege company-wide view, or a mix of the two.

• Read-only by default, with guarded writes. Changes are switched on deliberately, bulk changes are previewed first, and every action is attributed to the person who asked for it.

• Your own admin portal. Manage users, credentials, user privileges and connected apps, and see reporting on how AI is being used, all in one place.

• Deep knowledge of the systems we connect. We’ve spent a decade inside PSA, ITSM, CRM and RMM platforms. Our connectors understand how your instance is configured, translate internal codes into plain language, and stay well inside each platform’s API limits so they never slow down your other tools.

• Managed and supported. We maintain the connectors as vendors change their APIs, add new connectors every month, and provide UK-based support.

As Pacific IT Support’s President put it: “The team at Recursyv have been dealing with MSPs and service desk integrations for over 10 years so we had high trust from the get-go.”

Getting started

Recursyv AI starts at £99 / $139 per user per month (excluding VAT and sales tax). That includes up to three connected apps per user, your choice of Claude, Microsoft Copilot, ChatGPT or Gemini, managed setup, the admin portal, full security and UK-based support. Connectors are available across service desk and ITSM, CRM, RMM, knowledge, finance and Microsoft 365.

Book a 20-minute demo to see Recursyv AI in action. Then put it to work on your own systems and your own data with a free two-week trial, and see how much time your team could get back.

Book a 20-minute demo