The problem
Mainframes and legacy terminal applications were built for humans sitting at keyboards. They render full screens of text, expect precise keystroke sequences, and may use protocols such as TN3270. Dedicated terminal automation tools already support these protocols; an agent-operated Linux terminal client is another approach to evaluate.
Traditional automation approaches — screen scraping, macro recorders, custom scripts — are brittle, expensive to maintain, and break whenever the screen layout changes. Testing these systems is manual, slow, and requires specialized operators who are retiring.
How agend solves it
agend gives your AI agent a real terminal emulator running in a full Linux environment. The agent launches c3270 or any TN3270 client, and agend's MCP tools let it:
- See the screen — read the terminal output rendered by the client; validate any application-specific field semantics separately
- Send keystrokes — type credentials, navigate menus, submit transactions with raw key sequences (Tab, Enter, PF keys)
- React to output — wait for screen changes, detect error messages, validate transaction results
- Chain operations — login, navigate to CICS, run a transaction, verify the result, and log out — all in a single agent conversation
The agent doesn't need a special mainframe SDK. It operates the terminal emulator the same way a human operator would — the same tool, the same keystrokes, the same workflow.
What you can automate
- CICS transaction regression testing
- IMS/DB2 query validation after schema changes
- TSO/ISPF navigation and JCL submission
- AS/400 green-screen application testing
- VT100/VT220 terminal application workflows
- Any curses-based or ncurses TUI application