Agentic MCP server provisionning

Companies want to adopt AI agents but face problem, MCP servers are complex to set up.

Each API requires manual configuration, server provisioning, containerization, and infrastructure management.

Our intelligent agent eliminates manual MCP server provisioning entirely.

When a user or AI agent needs to access company APIs for a given task , my system:

Automatically discovers ALL necesarries APIs for a given task from natural language requests using RAG (Qdrant vector search, on swaggers)

Converts the Swagger specs to MCP servers on-the-fly using FastMCP

Provisions infrastructure on GCP with full automation (Terraform, Docker, Nginx)

Exposes production-ready endpoints in 3 minutes with zero manual intervention

Use this MCP to answer the request (This part indeed, has been not implemented yet)

Why This Matters:

Allow compay to develop agentic workflow that tinteract with their custom system thourgh API with agentic drag and drop tool WITHOUT having to develop MCP but rather utilitizing exsiting api

Scales infinitely - Upload new Swagger files to Qdrant, agent handles the rest, NO CONTEXT OVERLOAD

Removes DevOps bottleneck - No specialized knowledge needed to make APIs agent-accessible

Enables rapid AI adoption - Any existing Swagger spec becomes agent-ready immediately

Production-grade - Full infrastructure with health checks, monitoring, and auto-cleanup

Demo vs. Reality: While this demo shows a simple “get products” query, the architecture supports complex multi-API workflows (e.g., “analyze inventory, create purchase orders for low-stock items, schedule fulfillment, notify suppliers”) where the agent autonomously provisions and orchestrates 3 MCP servers to complete the task (use case not working at this point but whole pipeline implemented, just need few debugging)

Tech Stack: LangGraph (workflow), Mistral AI (reasoning), Qdrant (API discovery), FastMCP (server generation), Terraform (infrastructure), GCP (hosting)

Business Impact: Transforms hundred of API to MCP server translation task from a multi-day DevOps task to a 5-minute automated process, making enterprise AI agent adoption practical and scalable.

I have prior experience in building agent for API management.

I also work at LVMH to convert swagger to MCP so I was aware fast mcp as a functionnality for it. All the others idea (rag part, avoiding context overload, runtime discovery and instanciation of mcp sevrer directly by the agent on the fly at runtim is by me)