MCAP is a proposed protocol architecture for AI-driven legacy modernization. It integrates Anthropic’s Model Context Protocol (MCP) and Google’s Agent2Agent (A2A) to securely analyze COBOL systems using LLMs, while preserving business logic and enabling multi-agent collaboration with human oversight.
💡 MCAP = MCP (context-aware tool access) + A2A (agent orchestration) + mainframe security & domain extensions
- 🔒 Secure Mainframe Interface: Read-only access to sanitized COBOL, JCL, VSAM, and DB2
- 🧠 LLM Orchestration: Multi-agent system for logic extraction, code generation, validation
- 🔄 Context Graph Engine: Maintains semantic trace of legacy programs
- 👨💼 SME Validation Layer: Human-in-the-loop workflows and audit trails
- ☁️ Modern Output: Testable Java code, documentation, APIs
| Folder | Purpose |
|---|---|
architecture |
Architecture diagram + written description |
docs |
Full research paper and protocol documentation |
agents |
Description of each AI agent in the MCAP model |
security |
Compliance and data protection considerations |
prototype |
COBOL samples and LLM-generated outputs |
roadmap |
Milestones and community collaboration plan |
This is a research hypothesis, not a production-ready tool.
We invite collaborators from mainframe, AI, enterprise architecture, and modernization domains to validate and improve this framework.
📖 Full paper: index.md