Skip to content

AI Agents and MCP: Does Your Data Have an 'Identity Card'?

AI Semantic Identity Cover

"Will all our data problems be solved once MCP (Model Context Protocol) is here?"

Lately, the hottest topic among engineers is undoubtedly MCP and AI Agents. Anthropic’s release of MCP—a protocol for standardized AI access to local data and external tools—sent shockwaves through the industry. Yet, at the threshold of this new era, one fundamental challenge remains unsolved.

That challenge is the 'Reliability of Context.'

Giving an AI 'access' to data is entirely different from the AI truly 'understanding' the 'intent' behind that data. This article explores why structural modeling like Protobuf or Thrift is no longer enough in the age of MCP, and how giving data an 'Identity Card' can complete the intelligence of AI agents.

1. The Limit of MCP: "Able to Read, Unable to Know"

Suppose an AI agent has gained access to your database or codebase via MCP. It will encounter a sea of struct and message definitions.

// A message the AI agent might see through MCP
message UserState {
  int64 id = 1;
  string status = 2;
  int32 access_level = 3;
}

To the AI, this data is merely a collection of integers and strings. - "Is this a network packet? (Does it need transient processing?)" - "Is this a database row? (Does it require persistence management?)" - "Is this configuration set by a designer? (Is it the basis for business logic?)"

Though the AI has reached the data via the MCP 'highway,' it often falls into hallucinations because it doesn't know the 'role' the data plays within the system. Structure has been delivered, but Intent has not.

2. 'Intentional Modeling': Presenting Data's 'Identity Card' to AI

We believe that data should be perceived by machines as a 'Context for Intelligent Action,' not just a generic payload.

To achieve this, we replaced the single abstraction of struct with 'Declarative Keywords' that clarify the data's purpose at the syntax level.

  • record: Pure data transfer object (DTO)
  • entity: Core system state (DB Row)
  • table: Reference knowledge for business logic (Master Data)
// A declaration that allows AI to instantly recognize 'Knowledge'
table<ItemMaster> = { key: "id" } // "Ah, this is searchable reference data!"

entity PlayerAccount {
  1> [key] int64 accountId
} // "Ah, this is core data that requires state management!"

The moment data is given an 'Identity Card,' an AI agent reading it via MCP stops guessing. Just by looking at the keyword, it can independently decide, "This is an entity, so I must consider concurrency control and transactions," or "This is a table, so I should design a caching strategy." This is where the designer's intent is directly mapped to the AI's intelligence through syntax.

3. Unified AST: A Bridge Between Legacy and the AI Era

The global data infrastructure is already built on billions of lines of Protobuf and Thrift. Replacing this massive legacy all at once is impossible.

We solved this through the 'Unified AST (Abstract Syntax Tree)' hub. It allows you to maintain existing legacy systems as-is, while layering a 'Semantic Layer' on top that the AI can understand.

// Use existing Protobuf files without modification
include "legacy/inventory.proto"

// Simply tell the AI agent the 'Identity' of this data
table<inventory.Item> = { key: "item_id" }

Thanks to this architecture—where legacy and modern philosophies coexist—we can provide the richest and most accurate context to AI agents using MCP, without any breaking changes.

Conclusion: Data in the AI Era Must Be a 'Blueprint of Design'

In an era where AI agents autonomously judge and act, data modeling is no longer just a 'means of moving data.' It must be a 'Blueprint of Design' that prevents machines from getting lost.

Revealing the intent hidden within the structure and 부여ing a clear identity to data are the keys to unlocking the true performance of AI agents on the MCP highway.

Project DeukPack

A project born to restore 'Data Semantics' for the era of MCP and AI agents. DeukPack aims for a technical infrastructure where data's reason for existence is declared, allowing AI agents to operate as a perfect reflection of human design.