Internship - Study and Implementation of an MCP Layer (H/F)

This is a 3-month internship.

The objective of this internship is to study, design, and prototype the integration of a Model Context Protocol (MCP) layer on top of the iWE platform.

The MCP layer aims to expose the platform capabilities (data model, business operations, workflows) in a standardized way for LLM-based agents and external intelligent systems.

The final deliverable is a working Proof of Concept (PoC) demonstrating how iWE can be accessed and manipulated through MCP-compatible clients.

The iWE platform currently exposes a REST API to interact with dynamically defined data models.

With the emergence of LLM agents and AI-native tooling, a new integration paradigm is gaining traction:
instead of calling APIs directly, systems expose discoverable capabilities (tools, resources) via MCP.

This raises several technical challenges:

• How to map a dynamic, configuration-driven data model to MCP primitives

• How to expose CRUD and business operations as MCP tools

• How to ensure security, isolation, and access control in this new interaction model

• What API changes or abstractions are required to support MCP efficiently

A parallel internship will focus on the development of an MCP-based agent capable of interacting with the iWE platform. Both interns will work in close collaboration to ensure alignment between the MCP server design and the agent’s capabilities, with the objective of delivering fully compatible and coherent solutions.

The internship will be structured around exploration, design, and implementation:

• Study the iWE platform architecture

• Understand how data models and APIs are generated from configuration

• Analyze current REST API capabilities and limitations

• Study the Model Context Protocol specification and ecosystem

• Identify how iWE concepts map to MCP:

  • Entities → Resources

  • API operations → Tools

• Design an MCP exposure strategy for dynamic models

• Identify required changes to the existing REST API:

  • normalization of endpoints

  • improved introspection capabilities

  • metadata exposure

• Propose an abstraction layer if needed (adapter or gateway)

• Study MCP-specific security concerns:

  • authentication and identity propagation

  • authorization at tool/resource level depending on user profile

  • data isolation between tenants

• Propose and prototype a secure access model

• Implement a basic MCP server exposing part of the iWE platform

• Dynamically generate MCP tools from configuration

• Connect the MCP server to an existing iWE instance

• Test the MCP integration with an MCP-compatible client

• Evaluate usability and limitations

• Document findings and architectural recommendations

• API Design & Evolution: REST, dynamic schemas, introspection

• MCP / AI Integration: tools, resources, protocol lifecycle

• Backend Development: Node.js / TypeScript

• Security: authentication, authorization, multi-tenancy

• System Design: adapter vs native MCP server architecture

Key challenges include:

• Exposing a dynamic data model in a structured protocol

• Maintaining strict security guarantees in an agent-driven context

• Designing a scalable and maintainable integration layer

• A working MCP PoC server connected to iWE

• A technical study covering:

  • architecture choices

  • API evolution strategy

  • security model

• Recommendations for production integration

• Strong programming skills (JavaScript / TypeScript)

• Understanding of REST API design

• Interest in LLMs, AI agents, and emerging protocols

• Knowledge of software architecture and system design

• Familiarity with security concepts (auth, access control)