AI transformation for legacy systems: add AI to enterprise systems without a rewrite
Short answer: you don't have to replace a legacy system to make it useful to AI. Put a Model Context Protocol (MCP) server in front of its existing interface, and AI agents can discover and call its operations as tools. The system of record stays exactly as it is.
Why AI transformation stalls on legacy systems
Enterprise AI pilots usually start with a chatbot over documents. The next step, letting AI agents look up an order, check a policy, update a ticket or reconcile an account, needs access to the systems that actually run the business. Those are often:
- SOAP/WSDL web services built in the 2000s and still carrying critical traffic.
- ERP and CRM back ends with large, complex service catalogs.
- Mainframe and core platforms in banking, insurance, healthcare and government, usually reached through an integration layer.
- Relational databases and message queues where much of the business logic lives.
LLMs can't call these directly. The usual answer, a legacy modernization program that migrates or rewrites the system first, takes years and puts the AI roadmap on hold.
Three ways to connect a legacy system to AI
| Approach | Time | Risk | Trade-off |
|---|---|---|---|
| Rewrite or migrate, then add AI | Months to years | High | Clean result, but AI waits on the migration |
| Hand-written integration per AI use case | Weeks per use case | Medium | Duplicated glue code and inconsistent security |
| Generate an MCP layer over existing interfaces | Days | Low | Depends on the system exposing a usable interface |
How the MCP approach works
- Start from the contract you already have. A WSDL describes every operation and its XML types.
- Generate tools. Each operation becomes an MCP tool with a name, a description and a JSON Schema for its inputs. See how to connect a SOAP/WSDL service to an MCP server.
- Apply guardrails once. Read-only by default, writes only with explicit opt-in, input validation and audit logging in one place. See MCP server security: safe defaults.
- Connect any MCP-compatible client. Claude, ChatGPT, IDE assistants and custom agent frameworks can all use the same server.
What adding AI capabilities to a legacy system looks like
- A support agent that looks up customer and order status in a SOAP-based ERP.
- An operations assistant that answers questions from a core banking or policy administration system.
- An internal copilot that drafts updates for staff to approve, using operations that are explicitly opted in for writes.
Where legacy2mcp fits
legacy2mcp is an open-source (Apache 2.0) Python toolkit that automates this approach. It generates an MCP server from a WSDL today, with database and queue adapters planned, and an adapter interface for other systems. Install it with pip install legacy2mcp or see the GitHub repository.
FAQ
What does AI transformation mean for legacy systems?
Giving AI agents safe, structured access to the data and operations inside existing systems of record, so they can answer questions and complete work using them.
Do I need to modernize a legacy system before adding AI?
No. An MCP server sits in front of the existing interface. You can still modernize later, and the agent-facing tools can stay the same.
Which enterprise systems can be connected to AI this way?
Any system with a machine-readable interface. SOAP/WSDL is supported today, which covers many ERP, CRM, banking, insurance and mainframe integration layers.
Is it safe to connect AI agents to systems of record?
It can be, with guardrails: read-only defaults, opt-in writes, schema validation, least-privilege credentials and audit logs.