The main difference is that term-cli is intentionally not MCP. There's been a shift away from MCP, to the more token-efficient combination of CLI tools plus agent skills. Microsoft explains it well here: https://github.com/microsoft/playwright-cli
Beyond that, term-cli focuses on agent-centric features like file transfer, full mouse support, color annotations, joint agent+human sessions (via term-assist) and more. I had Claude Opus drive the interactive colored menus of OpenCode (TUI) while reproducing a bug.
term-cli is designed to meet agents' needs, and not just wrapping a terminal. Another example: humans click on coordinates, but agents are really bad at counting terminal columns due to tokenization. So with term-cli, agents instead define the string they want to click on.
how's this different from the many existing tmux/terminal MCP servers?
The main difference is that term-cli is intentionally not MCP. There's been a shift away from MCP, to the more token-efficient combination of CLI tools plus agent skills. Microsoft explains it well here: https://github.com/microsoft/playwright-cli
Beyond that, term-cli focuses on agent-centric features like file transfer, full mouse support, color annotations, joint agent+human sessions (via term-assist) and more. I had Claude Opus drive the interactive colored menus of OpenCode (TUI) while reproducing a bug.
term-cli is designed to meet agents' needs, and not just wrapping a terminal. Another example: humans click on coordinates, but agents are really bad at counting terminal columns due to tokenization. So with term-cli, agents instead define the string they want to click on.