What is MCP?
Model Context Protocol (MCP) is an open standard that defines how AI models connect to external tools and data sources. Think of it as a universal plug for AI. Instead of building custom integrations between every AI model and every business tool, MCP provides a single, consistent way for them to talk to each other.
Before MCP, if you wanted an AI assistant to look something up in your CRM, check your calendar, and draft an email, each of those connections had to be built separately. Different APIs, different authentication flows, different data formats. It worked, but it was expensive and fragile.
MCP changes that. It defines a standard protocol that any AI model can use to discover what tools are available, understand what they do, and call them with the right parameters. One pattern for everything.
Why it matters for business
Here's the practical version. Right now, most businesses using AI have it operating in a bubble. The AI can answer questions based on what it was trained on, or maybe what's in a specific document set you've connected via RAG. But it can't reach into your live systems to check an order status, look up a customer record, or update a ticket.
MCP is what closes that gap. With MCP-compatible tools:
- An AI assistant can query your Xero or MYOB instance to answer a finance question
- A customer service AI can look up order history, check shipping status, and update case notes, all in one conversation.
- An internal knowledge agent can search across SharePoint, Confluence, your database, and your email system simultaneously
- An AI can actually take actions (create a draft invoice, schedule a meeting, file a ticket), not just answer questions about them.
The shift is from AI that talks about your business to AI that works inside your business. That's a significant difference in value.
How MCP works, without the jargon
MCP uses a client-server model. The AI application (like Claude, ChatGPT, or your custom AI assistant) is the client. Your business tools and data sources each run a small MCP server that exposes their capabilities.
When the AI starts up, it asks: "What tools are available?" Each MCP server responds with a list of things it can do: search records, create entries, pull reports, whatever makes sense for that tool. The AI then knows what's available and can call those tools when it needs to.
Three core concepts:
- Tools. Actions the AI can take. "Search Xero contacts," "Create a Jira ticket," "Look up a patient record." Each tool has a description and defined inputs/outputs.
- Resources. Data the AI can read. Documents, database records, configuration files. Like giving the AI access to a filing cabinet.
- Prompts. Predefined templates for common tasks. "Summarise this customer's last 5 interactions" or "Generate a monthly revenue comparison." Structured workflows the AI can follow.
The clever part is that the AI doesn't need to know the internals of each system. It just needs to know what tools are available and what they accept. The MCP server handles the actual integration (authentication, data transformation, error handling) all behind a consistent interface.
MCP vs custom API integrations
You might be thinking: "We already have API integrations. How is this different?"
Fair question. Traditional integrations connect System A to System B with custom code. They're point-to-point. If you have 10 systems and you want them all connected, that's potentially dozens of individual integrations to build and maintain.
MCP changes the geometry. Each system publishes its capabilities through one MCP server. Any MCP-compatible AI client can then use any of those tools. You build the MCP server once, and every AI application can use it.
| Aspect | Custom API Integration | MCP |
|---|---|---|
| Connection pattern | Point-to-point | Hub-and-spoke |
| New AI model | Rebuild each integration | Works automatically |
| Discovery | Hardcoded in each app | AI discovers tools dynamically |
| Maintenance | Each integration maintained separately | One server per data source |
| Best for | Deterministic system-to-system flows | AI-driven, flexible interactions |
That said, MCP doesn't replace all integrations. If you need Xero to sync with your inventory system on a schedule, that's still a traditional integration job. MCP is specifically about giving AI models access to tools and data. The two approaches complement each other.
Who's behind it
Anthropic created MCP and open-sourced it in late 2024. That's significant because Anthropic isn't a small player. They build Claude, one of the most capable AI models available.
What turned MCP from "interesting Anthropic project" to "industry standard" is who adopted it afterwards. By early 2026:
- OpenAI added MCP support to ChatGPT and their API
- Google added MCP compatibility to Gemini
- Microsoft integrated MCP into Copilot and Azure AI
- Tool vendors like Slack, GitHub, Notion, Atlassian, Salesforce, and dozens more published official MCP servers
When every major AI provider and most business tool vendors support the same protocol, it's no longer optional. It's infrastructure. Similar to how REST APIs became the standard for web services. Once enough players adopt it, everyone follows.
Real-world use cases
This is where it gets practical for Australian businesses:
Internal knowledge assistant
A law firm connects MCP servers to their document management system, practice management software, and billing system. Staff ask the AI questions like "Show me all matters for Client X that are overdue" and get accurate answers pulled from live data, not from a stale export or a document that's three months old.
Customer service automation
A trades business connects their job management system (ServiceM8, Simpro) via MCP. The AI assistant can look up job status, check technician availability, and book follow-up appointments, all within the same customer conversation. No more "let me check and get back to you."
Finance and operations
An operations manager asks their AI assistant: "What were our top 10 customers by revenue last quarter, and which ones have outstanding invoices?" The AI queries Xero through MCP, cross-references with the CRM, and returns a consolidated answer. That used to be a 30-minute spreadsheet exercise.
AI-powered document processing
An insurance company processes claims by connecting their claims system, policy database, and document store via MCP. The AI reads the claim, pulls the relevant policy, checks coverage, and flags anything that needs human review. The processing time drops from hours to minutes.
What MCP doesn't solve
MCP is promising, but it's not magic. A few things worth understanding:
Security is still your problem. MCP makes it easy for AI to access your tools. That's powerful and risky. You need clear policies about which tools the AI can use, what data it can access, and what actions it can take autonomously versus what requires human approval. The protocol supports permissions, but you have to configure them properly.
Not every tool has an MCP server yet. The major platforms are covered, but niche Australian software (industry-specific tools, legacy systems, custom databases) probably won't have off-the-shelf MCP servers. You'll need custom ones built. That's not hard, but it's not free either.
Quality depends on the underlying AI. MCP gives the AI access to better information, but it doesn't make the AI smarter. If the model hallucinates or misinterprets a query, having access to live data can actually make things worse, because the hallucination now looks more credible. Human oversight still matters.
It's still maturing. The spec is evolving. Authentication patterns, error handling, and governance features are improving with each version. Early adopters need to stay on top of updates.
What to do about it now
If you're running a business and wondering whether MCP matters to you, here's the practical filter:
- Already using AI? Check whether your AI tools support MCP. If they do, you can start connecting business systems without building custom integrations from scratch.
- Planning an AI project? Make sure your architecture accounts for MCP. Any new AI system should be built to use MCP-compatible tool connections rather than bespoke integrations that'll need replacing in 12 months.
- Not using AI yet? MCP lowers the barrier. Previously, connecting AI to your actual business data was a major integration project. With MCP, it's becoming more like configuring a connection than building one.
The key takeaway: MCP is becoming the standard way AI connects to business systems. It doesn't change what AI can think. It changes what AI can do. And for most businesses, that's the part that actually matters.
Key takeaways
- MCP is an open protocol that standardises how AI models connect to external tools, databases, and business systems.
- It means AI assistants can read your CRM, search your documents, and take actions, all through a single, consistent interface.
- Anthropic created it, but Google, OpenAI, Microsoft, and dozens of tool vendors now support it.
- For businesses, MCP reduces the cost and complexity of integrating AI with the systems you already use.
