- 11
- March
MCP (Model Context Protocol) is an open standard developed by Anthropic for systematically connecting AI systems to external tools and data sources — like "USB for AI" that allows AI applications to read files, call functions, and access data from various systems through a unified interface, without writing separate integration code for each one.
In Brief: What Is MCP?
MCP = a universal standard that lets AI communicate with external tools, just like USB lets computers connect to any device through a single port. Before MCP, developers had to write separate integration code for each system (N x M integrations), but MCP reduces this to write once, use everywhere (N + M integrations).
Key Timeline of MCP
MCP is a relatively new technology, but it has evolved remarkably fast:
| Period | Key Event |
|---|---|
| November 2024 | Anthropic launched MCP as an open standard available free for developers worldwide. |
| Early 2025 | Major tech companies began supporting MCP, including Block, Replit, Sourcegraph, and Zed. |
| December 2025 | Anthropic transferred MCP to the Agentic AI Foundation (AAIF) under the Linux Foundation, alongside Block and OpenAI, to become an industry standard. |
| Early 2026 | Claude has 75+ connectors via MCP, supporting connections to numerous popular services. |
How Does MCP Architecture Work?
MCP uses a Client-Server architecture consisting of 3 main components:
| Component | Role | Example |
|---|---|---|
| MCP Host | The AI application that users interact with | Claude Desktop, IDE, enterprise chatbot |
| MCP Client | Middleware that creates 1:1 connections with MCP Servers | SDK embedded in the Host application |
| MCP Server | Lightweight programs that expose specific data or tools | Google Drive Server, Database Server, ERP Server |
When a user asks a question, the AI determines which data source is needed and automatically calls the relevant MCP Server via the MCP Client. All of this happens behind the scenes — the user doesn't need to do anything extra.
Key Features of MCP
MCP is not just a simple API connection — it has capabilities specifically designed for AI:
| Feature | Details |
|---|---|
| Tool Calling | AI can directly call functions from external systems, such as creating purchase orders, sending emails, or pulling reports. |
| Resource Access | AI can read files, databases, or documents from various data sources without copy-pasting. |
| Contextual Prompts | MCP Servers can send prompt templates to AI for better understanding of data context. |
| Tool Search | AI can automatically discover appropriate tools from a registry without pre-configuration. |
| Programmatic Tool Calling | Call tools directly via API without going through the AI's UI. |
| Sampling (Requesting AI Assistance) | MCP Servers can request AI to help process data — enabling 2-way communication. |
Comparison: Before and After MCP
To illustrate the difference, let's compare traditional AI-to-external-system integration with MCP:
| Aspect | Before MCP | After MCP |
|---|---|---|
| Integration | Write separate code for each system (Custom Integration) | Write one MCP Server, use with any AI |
| Number of Integrations | N AI apps x M systems = N x M pieces of work | N + M pieces of work (each side writes only once) |
| Standards | No universal standard; each vendor does their own thing | Open standard under the Linux Foundation |
| Security | Depends on each developer | Built-in permission model; users must approve before AI can access |
| Maintenance | Update each integration individually when APIs change | Only update the MCP Server that changed |
| Vendor Lock-in | Locked to a single AI provider | Switch AI providers without rewriting integrations |
Why Is It Called "USB for AI"?
Before USB, computers needed separate ports for each device (Serial, Parallel, PS/2) — USB made it possible to connect everything through one port. MCP does the same for AI — instead of writing separate integration code for each system, MCP provides a single interface that any AI can use to connect to any system.
Use Cases: How Is MCP Used in Practice?
MCP is already being used in production across various scenarios, from developer-level to enterprise-level:
1. AI Assistants That Access Enterprise Data
Instead of copy-pasting data from various systems for AI to read, MCP lets AI directly access databases, Google Drive, Slack, or ERP systems. For example, asking "How do this month's sales compare to last month?" and the AI pulls data from the sales database to analyze instantly.
2. Automated Software Development
Developers use MCP to let AI access the codebase, Git repository, CI/CD pipeline, and documentation simultaneously, enabling AI to help write code, debug, and deploy with real project context understanding.
3. Connecting Business Tools
Connect AI to Google Calendar, Gmail, CRM, document management systems, and more via MCP, enabling AI to help manage daily tasks such as summarizing important emails, scheduling meetings, or generating reports automatically.
4. Real-time Data Analysis
AI connects to databases via MCP Servers and can query data for analysis instantly — no more exporting to CSV and uploading to AI. This aligns with the concept of seamless system integration (ERP Interoperability).
5. Agentic Workflows
MCP is a critical foundation for AI Agents that can automate multi-step workflows. For example, the command "create a weekly sales summary report and send it to the management team" — the AI pulls data from the database, creates the report, and sends the email automatically, all through MCP.
Security Considerations
- Permission Control: Define permissions for each MCP Server clearly — don't grant access to everything.
- Data Leakage: AI may send sensitive data to the LLM provider if the scope of data exposed by MCP Servers is not restricted.
- Prompt Injection: Data from external sources may contain malicious prompts — a data filtering system is needed before sending to AI.
- Audit Trail: Log every MCP Server call for traceability, in line with AI Governance principles.
- Server Authentication: Verify that connected MCP Servers are authentic — not fake servers that could steal data — following the same principles as enterprise cybersecurity.
MCP and ERP: New Opportunities for Organizations
MCP opens new opportunities for enterprise ERP systems to connect with AI much more easily:
- AI that understands ERP data: Instead of navigating ERP screens to find reports, users can ask AI and get answers instantly.
- Workflow Automation: AI creates Purchase Orders, approves documents, or generates reports via MCP Servers connected to ERP.
- Cross-system Intelligence: AI pulls data from ERP, CRM, and warehouse systems simultaneously for 360-degree analysis.
Saeree ERP and AI
Saeree ERP is developing an AI Assistant, currently in the Training phase to help AI understand the context of Thai enterprise ERP data. MCP technology is one of the approaches the development team is studying to enable the AI Assistant to efficiently connect with various Saeree ERP modules in the future.
Who Is MCP For?
| Suitable For | Not Yet Necessary For |
|---|---|
| Organizations already using AI in daily operations | Organizations that haven't started using AI yet |
| IT teams that need to connect AI to multiple internal systems | Organizations using AI only for general Q&A chat |
| Developers building AI-powered applications | General users who don't develop software |
| Organizations wanting to reduce vendor lock-in with AI providers | Organizations using a single AI provider and satisfied |
| Businesses planning a long-term AI strategy | Businesses whose digital data isn't ready yet |
MCP transforms how AI connects to the outside world — from writing custom code for each system to an open standard everyone can share. Organizations that understand and prepare for MCP today will have an advantage in leveraging AI to its full potential in the future.
- Saeree ERP Team
Summary
MCP (Model Context Protocol) is an open standard that enables AI to systematically connect with external tools and data sources. Developed by Anthropic, it is now governed by the Linux Foundation through the Agentic AI Foundation.
What organizations should do now:
- Follow MCP developments — as it's becoming an industry standard.
- Assess systems that need AI connectivity — ERP, CRM, databases, documents.
- Plan your AI Strategy — aligned with AI Governance principles to ensure safe and transparent AI use in the organization.
- Choose an AI-ready ERP system — systems with open APIs and external system integration capabilities will have an advantage in the AI era.
