Written by: Mikaela Crimmins - Chief Strategy Officer at Orchard

Mikaela Crimmins, Chief Strategy Officer at Orchard, talks to Martin Stafford, COO about the small but important thing holding most AI projects back.

What is Model Context Protocol (MCP)?

Model Context Protocol (MCP) is an open standard that lets AI agents connect to and act within business systems using authenticated, audited, user-level access. It is backed by Anthropic, OpenAI, Google, and Microsoft. Where most AI tools can only read data, MCP-connected agents can take action inside the systems organisations actually run on.

Mikaela: Most organisations are looking at ways to automate systems, why or where do you think it stalls? 

Martin: It almost always comes down to coverage. The big platforms have done real work on connectors. Microsoft 365, HubSpot, Salesforce, Atlassian. If your whole operation runs on those, you can connect AI to a decent chunk of your business. But most organisations aren't like that. They've got those core platforms, and then all the other stuff. Bespoke builds. Legacy systems. Industry-specific tools that have been around for years. Those systems are usually where the important operational data actually lives. And right now, AI can't see any of it.

The number that matters: you need about 90 to 95 percent of your systems connected to automate a full process end to end. Most organisations are at about 60 percent. This is usually the frustrating part and all of the ‘good data’ or company IP is sitting in places that don’t have ready-made connectors.

“Most organisations have connected around 60% of their systems to AI. Automating a full process end to end requires 90–95%.”

Martin Stafford, COO, Orchard

Mikaela: So what helps bridge that gap?

Martin: MCP is Model Context Protocol. Open standard, backed by Anthropic, OpenAI, Google, and Microsoft. It's basically a way to connect an AI agent to any system you need it to talk to. The concept has been around for a while, but until recently it wasn't ready for real business use. No proper security, shared credentials, no audit trail. Not something you'd put in front of a compliance team.

What changed is they added proper authentication and streaming. Now the AI acts on behalf of a specific person, using that person's own permissions. It can only do what you can do. Every action gets logged. That took it from a prototype to something you'd actually put into production.

Mikaela: When you build one of these for a client, is it as simple as just plugging two things together?

Martin: A basic connector is just a wrapper around an API. The AI can call it. That's useful, but it's the starting point. We go further than that. The connector is where you embed your business logic. Who gets read access, who gets write. You turn complex multi-step operations into a single action so the AI doesn't have to guess its way through a process. And you define what the data actually means in context. Which data is sensitive. Which actions can't be undone. What a record actually represents.

This is critical, as you’re not just handing over the keys, you’re setting the rules of access.

Mikaela: So where can Orchard help organisations?

Martin: We focus on the systems where a ready-made connector doesn't exist and probably won't for years. The business that built the software has no commercial reason to make one, but the companies using it have real value locked inside. For example we've built a connector for Deltek WorkBook, an agency management platform that handles everything from project scoping through to billing and financial close. Operational backbone for a lot of agencies, no AI integration path. We also built one for Veeva Promomats, which manages regulatory submissions for pharma promotional content. Same story. Both sit right at the heart of how those businesses run. And a well-scoped connector is weeks of work, not years. The hard part isn't the build. It's knowing exactly what you need it to do, and for these organisations the impact is significant.

Mikaela: If someone's reading this and thinking ‘that sounds like us’, where do they start?

Martin: Work out what you're actually trying to automate. Then list every system that process touches. Some of those will already have connectors. Some won't. That's your gap, and it's usually smaller than people think, maybe one or two systems. But those one or two are the reason the whole thing doesn't work. We can build those connectors. It's a small piece of work, but it's the piece that makes everything else click

Common questions

What is Model Context Protocol (MCP)?

MCP is an open standard that enables AI agents to connect to, read from, and take action within business systems. It uses authenticated, user-level access so AI can only do what the authorised person can do, and every action is logged. It is backed by Anthropic, OpenAI, Google, and Microsoft.

Why do enterprise AI projects stall?

The most common reason is incomplete system connectivity. To automate a business process end to end, AI needs access to roughly 90–95% of the systems that process touches. Most organisations have connected around 60%. The remaining gap, often one or two bespoke or industry-specific platforms, is what prevents the whole thing from working.

What does an MCP connector actually do?

A connector is a secure bridge between an AI agent and a business system. It defines access permissions, embeds business logic, translates complex multi-step operations into single actions the AI can call, and specifies which data is sensitive and which actions are irreversible. It is the layer where governance and safety are built in.

How long does it take to build a custom MCP connector?

A well-scoped connector typically takes weeks, not months. The build itself is not the hard part. The work is in scoping precisely what the connector needs to do, what data it needs to access, and what rules govern its behaviour. Orchard has built production connectors for Deltek WorkBook and Veeva Promomats.

Which systems does Orchard build MCP connectors for?

Orchard focuses on systems where a ready-made connector does not exist and is unlikely to be built by the platform vendor. These are typically bespoke builds, legacy platforms, or industry-specific tools where the operational data is valuable but inaccessible to AI. Current examples include Deltek WorkBook (agency project and financial management) and Veeva Promomats (pharma regulatory submissions).

Mikaela Crimmins is Chief Strategy Officer at Orchard, a connected experience agency with specialists across Sydney, Hobart, and New York. She leads strategic planning for enterprise clients across financial services, healthcare, and consumer sectors.

Martin Stafford is COO and CTO at Orchard. He has led AI integration architecture for enterprise clients across financial services, aged care, pharma, and professional services, and is the architect behind Orchard’s MCP connector program.