Romain Cadarsi

Design and implementation of an MCP overlay for Evolut, a multi-site Symfony CMS, to open the application to AI agents capable of interacting with the CMS in natural language, without direct database access.

The project first involved creating a secure bridge between Evolut and AI agents: exposing dedicated MCP tools, authenticated access to business data, reading and modifying pages, blocks, CustomPages, CustomPageModels, global parameters, SEO content, images, and publication statuses. This layer allows an agent to understand the CMS structure, act on the right objects, and verify its actions through HTML rendering, previews, data rereading, and draft/publication control.

Once this MCP overlay was stabilized, I encapsulated the main Evolut operational workflows into reusable AI skills. Each skill describes a complete work method: business context, security rules, MCP tools to use, expected payloads, verification steps, CMS-specific best practices, and limits to respect.

The goal was to decouple the agent's performance from the AI solution used. Business knowledge, procedures, and safeguards no longer depend solely on an agent or a prompt: they are carried by a structured MCP layer and transferable skills, making it easier to interchange AI agent solutions without starting from scratch or losing execution quality.

This mission transformed Evolut into an AI-augmented CMS, capable of accelerating the creation, maintenance, audit, SEO optimization, and adaptation of websites while respecting the product's technical structure, access rights, and business rules.

Development of an autonomous AI agent based on OpenClaw, deployed on a server, to automate the monthly SEO monitoring of sites managed with Evolut.

The initial need was to monitor the search performance of a growing client base, without increasing manual tasks for the team. As Evolut sites are connected to Google Search Console and Matomo, I designed an agent capable of consuming the Evolut MCP, retrieving Search Console data, cross-referencing it with Matomo statistics via a dedicated MCP server, and then generating a PDF-exportable SEO report at the beginning of each month.

The agent relies on a manually built knowledge base to interpret signals: trending queries, pages viewed, underperforming content, editorial opportunities, and priority actions to implement.

I also integrated a complete email workflow: reading client responses, maintaining stable context per project, preparing drafts for validation, sending reports, and automated exchanges. Based on the previous month's data, the agent suggests 10 relevant blog topics, then automatically prepares the selected content according to project criteria: word count, structure, summary, editorial tone, and SEO constraints.

Result: over 150 clients can be informed each month about their website's search performance evolution, with actionable recommendations and regular SEO content production.

Subcontracting mission for CopyCatGroup, a company specializing in printing and document services, acting as a versatile technical provider for various aspects of their digital business.

I am involved in maintaining existing tools, creating websites, and developing AI solutions to streamline internal processes and services offered to their clients. The goal is to provide reliable technical support that covers both classic web needs and advanced automation.

A significant part of the mission involves creating autonomous AI agents for managing client social media: building editorial calendars, preparing content, generating or preparing visuals, and then publishing directly on multiple social platforms.

This collaboration combines application maintenance, web development, AI automation, multi-tool orchestration, and business workflow optimization.