In recent years, the adoption of artificial intelligence in business environments has grown exponentially. Next-generation language models (LLMs) like GPT, Claude, or Llama have demonstrated their ability to automate tasks and deliver sophisticated conversational experiences. However, a significant challenge remains: connecting these models to real services to ensure responses are useful, up-to-date, and secure.
Limitations of the Function Calling Approach
Traditionally, integrations were implemented through "function calling," where each LLM defines its own format for invoking external APIs. This approach presents issues such as:
Lack of standardization: Each model uses different JSON structures.
Low predictability: Models may misinterpret functions or arguments.
Fragile connectors: Each API requires a custom implementation.
The result: a fragmented ecosystem that complicates scalability and maintenance.
MCP: A New Era of Interoperability
The Model Context Protocol (MCP) emerges as an open standard that separates the model's decision logic (what to do) from execution (how to do it), through a common protocol based on JSON-RPC.
Why MCP Makes a Difference
Universal protocol: MCP allows clients like Jelou to interact with multiple services regardless of the underlying model.
Reduced friction: A common language is used for calls and responses, avoiding custom logic.
High scalability: Simply add a new MCP server to integrate a system, without rewriting prompts or integrations.
"MCP offers a way to universalize the execution of model-generated instructions, maintaining compatibility and simplifying architecture."
Jelou as an MCP Host
At Jelou, we believe the key to the future lies in facilitating secure and agile connections between AI and real services. Therefore, we are working to make our platform function as an MCP Host, allowing conversational assistants to access external data and functionalities in a standardized way.
Additionally, we promote a development environment that enables:
Building custom MCP Servers within Jelou.
Easily configuring business logic with the help of an LLM.
Maintaining security by isolating each service in a secure environment.
Comparison: Function Calling vs. MCP
| Approach | Standardization | Interoperability | Scalability |
|---|---|---|---|
| Function Calling | Low | Limited | Fragile |
| MCP (with JSON-RPC) | High | Universal | High |
Practical Example: Client Search in CRM
Function Calling (GPT-4o)
functions = [{
"name": "search_client",
"description": "Searches for client information in the CRM",
"parameters": {
"type": "object",
"properties": {
"email": {"type": "string"},
"fields": {"type": "array", "items": {"type": "string"}}
}
}
}]MCP (JSON-RPC)
{
"jsonrpc": "2.0",
"method": "crm.searchClient",
"params": {
"email": "juan@company.com",
"fields": ["name", "last_order"]
},
"id": 1
}Bridge Architecture: Connecting Non-Compatible Models
For models that do not yet natively support MCP, at Jelou we are developing a bridge component called Bridge. This translates calls from traditional models to MCP, ensuring full interoperability.
The Model Context Protocol represents a key advancement in connecting AI with real services in a standardized and scalable manner. At Jelou, we are committed to more powerful, accessible, and reliable AI, and we are building the foundations for any company to easily integrate into the conversational future.
