MCP Is Becoming the Plumbing of Agentic ERP: 2026 Guide
MCP (Model Context Protocol) is quietly becoming the plumbing of agentic ERP, according to a new ERP Today analysis. If that headline sounds like alphabet soup, don't worry — here's what actually happened, why it matters even if you never touch an ERP system, and how you can try MCP yourself today.
📰 What Happened: ERP Today Says MCP Is the New Plumbing of Business Software
ERP Today, a trade publication covering enterprise software, published an analysis arguing that MCP — the Model Context Protocol — is becoming the standard 'plumbing' that connects AI agents to ERP systems (the big software platforms companies use to run finance, inventory, HR, and operations). The piece urges ERP leaders to understand MCP now, because it's the layer that will decide how AI agents actually read and write business data.
Quick translation: MCP is an open standard, originally introduced by Anthropic in November 2024, that gives AI models a common way to connect to outside tools and data — think of it as a USB-C port for AI. Instead of every software vendor building a custom integration for every AI model, everyone plugs into one shared protocol.
The reason this is news in 2026 is momentum. MCP went from an Anthropic project to an industry-wide standard: OpenAI announced support in 2025, Google and Microsoft followed, and major enterprise vendors have been shipping MCP-compatible connectors. When the trade press starts calling something 'plumbing,' it means the debate is over — it's infrastructure now.
🔌 Wait — What Are 'Agentic ERP' and MCP, in Plain English?
'Agentic ERP' just means business software where AI agents do the work instead of humans clicking through menus. Instead of you opening the accounting module, finding an invoice, and approving it, you tell an AI agent 'approve all invoices under $500 from verified vendors' and it does the clicking — safely, with logs and permissions.
For that to work, the AI needs a reliable way to actually reach the data and take actions. That's what MCP provides. An 'MCP server' is a small connector that exposes a system's data and actions (like 'look up an order' or 'create an invoice') in a format any MCP-compatible AI can use — whether that AI is Claude (e.g. Claude Sonnet 4.6), OpenAI's GPT-4o, or Google's Gemini 2.0.
The plumbing metaphor is apt: you don't admire plumbing, and you rarely think about it. But nothing in the house works without it. The article's argument is that whoever controls or understands this connective layer controls how useful AI agents become inside a business.
The USB-C Analogy
Before USB-C, every device had its own charger. Before MCP, every AI integration was custom-built: one connector for Claude-to-Salesforce, another for ChatGPT-to-SAP, and so on — an M×N explosion of integrations. MCP collapses that into one plug shape. Build one MCP server for your system, and every MCP-compatible AI can use it.
💼 Why This Matters for Solopreneurs and Knowledge Workers
You probably don't run SAP. So why care? Because the same plumbing being installed in giant ERP systems is already available in the tools you use every day. Claude, ChatGPT, and other assistants can connect via MCP to Notion, Google Drive, Slack, GitHub, Stripe, Shopify, and hundreds of other services. The ERP story is the enterprise-scale proof that this approach won.
Second, it changes what 'using AI' means. In 2023–2024, AI was a chat window you copy-pasted into. In 2026, AI is becoming a worker that reaches into your actual systems: it reads the real invoice, updates the real spreadsheet, and files the real ticket. MCP is the reason it can do that across different apps without you buying a special integration for each pair.
Third, there's a career and business angle. When a standard becomes plumbing, the people who understand it early have an advantage. Solopreneurs who wire their small stack (say, Shopify + Gmail + Notion) into an AI agent can automate work that used to require hiring. Consultants and freelancers who can say 'I set up MCP-based agent workflows' are selling a skill that mid-size businesses are just starting to want.
⚖️ Old Integrations vs. MCP: What Actually Changed
It helps to see the before-and-after side by side. The old world wasn't broken — it was just expensive and slow. Every connection between an AI and a business system was a bespoke project. MCP standardizes the connection itself, so the effort shifts from 'building pipes' to 'deciding what flows through them.'
That last part is the real message to ERP leaders in the article's framing: the plumbing is arriving whether you plan for it or not. The remaining questions are governance ones — which data agents can touch, what actions need human approval, and how everything is logged.
| Aspect | Old Way (Custom Integrations) | MCP Way (2026) |
|---|---|---|
| Connections needed | One per AI–app pair (M×N) | One MCP server per app (M+N) |
| Who builds it | Developers, per project | Vendor ships it once; anyone connects |
| Switching AI models | Rebuild the integration | Same connector works with Claude, GPT-4o, Gemini 2.0 |
| What the AI can do | Read-only or narrow tasks | Read data AND take approved actions |
| Main risk | Integration breaks or goes stale | Governance: permissions, logging, oversight |
🚀 How You Can Try MCP Today (No Code Required)
You don't need an ERP system or a developer to experience what this article is describing. The consumer versions of major AI assistants already use MCP under the hood for their 'connectors' or 'integrations' features.
The easiest entry point: open Claude (web or desktop) and look at Settings → Connectors. You can link tools like Google Drive, Notion, Slack, or Canva, then ask Claude to do real work across them — 'find my Q2 pricing doc and summarize what changed.' That connection is MCP in action. ChatGPT offers a similar path through its connectors and developer mode, and desktop apps like Claude Desktop let you add community-built MCP servers for hundreds of other tools.
If you're slightly technical, the next step up is browsing the open-source MCP server directory on GitHub (modelcontextprotocol/servers) — there are connectors for databases, CRMs, e-commerce platforms, and file systems. And if you run a business on Shopify, Stripe, or similar platforms, check their docs: many now publish official MCP servers you can plug an AI agent into directly.
- ✔Open Claude or ChatGPT and find the Connectors/Integrations settings
- ✔Connect one low-risk tool first (e.g. Google Drive or Notion)
- ✔Ask the AI to complete a real task across that connection
- ✔Check what permissions you granted — practice the governance habit early
- ✔Browse github.com/modelcontextprotocol/servers to see what else exists
- ✔If you sell services, note which client tools already have MCP connectors
🔭 What to Watch Next: The Governance Question
The ERP Today framing — 'leaders need to understand it' — is really a warning about governance, and it applies at every scale, including a one-person business. Once an AI agent can take actions in your systems, the important questions stop being about capability and start being about control: What can it touch? What requires your approval? What gets logged?
Enterprises are solving this with permission scopes, audit trails, and human-in-the-loop approval steps. As a solopreneur, you can copy the same pattern for free: give agents read access before write access, keep approval steps on anything involving money or customers, and review what your connected tools can see.
Expect the next 12 months to bring more official MCP servers from major SaaS vendors, better security tooling around agent permissions, and — as the article suggests — ERP vendors competing on how well their systems expose data to agents rather than on their own built-in AI features. The plumbing is being installed now; the interesting question is what gets built on top of it.
❓ Frequently Asked Questions
What does MCP stand for and who created it?
MCP stands for Model Context Protocol. It's an open standard introduced by Anthropic (the company behind Claude) in November 2024. It defines a common way for AI models to connect to external tools, data sources, and applications. Since then it has been adopted well beyond Anthropic — OpenAI, Google, Microsoft, and many enterprise software vendors support it.
Is MCP only for big companies with ERP systems?
No. ERP is just where the enterprise money is, which is why trade publications cover it. The same protocol powers the connectors in consumer AI tools — linking Claude or ChatGPT to Google Drive, Notion, Slack, Shopify, and more. A solopreneur can use MCP-based connections today without writing any code.
Is it safe to let an AI agent access my business data through MCP?
It's as safe as the permissions you grant. MCP itself is just the connection standard; safety comes from scoping access (read-only vs. write), using official connectors from trusted vendors, and keeping human approval on sensitive actions like payments. Start with low-risk tools, review permissions, and expand gradually.
What's the difference between an AI agent and a regular chatbot?
A chatbot answers questions in a conversation window. An agent can take actions in other systems — searching your files, updating a record, creating an invoice — usually through connections like MCP. Models such as Claude Sonnet 4.6, GPT-4o, and Gemini 2.0 can operate in both modes; the agent behavior comes from what they're connected to.
🏁 Final Thoughts
The short version: MCP has won the standards race for connecting AI agents to business software, and ERP Today's 'plumbing' framing makes it official — this is infrastructure now, not an experiment. You don't need to run an ERP system to benefit. The same protocol is sitting inside the AI tools you already use, waiting for you to connect your first app. Try one connector this week, keep your permissions tight, and you'll understand agentic AI better than most executives reading that headline. If this explainer helped, subscribe to Agents at Work for weekly plain-English breakdowns of AI news — and drop a comment telling us which tool you'd connect an AI agent to first.
Last updated: July 07, 2026 · Keyword: MCP agentic ERP · Agents at Work
Comments
Post a Comment