Why a Grok Bot hooks into t3 code via CLI instead of MCP
Two small CLIs, t3ctl and t3listen, connect the author's Grok Bot to t3 code's local server because MCP lacks push notifications, can't handle permission prompts, and wastes limits on polling.
a lot of people asked why not just use t3 code's mcp? here's why right now each of my computers runs t3 code and the grok bot app. tots adds two small clis there, t3ctl and t3listen, that talk to t3 code's local server over its websocket api why not mcp yet: > no push/notif. mcp doesn't tell/notify my bot when a thread asks something or finishes, so the bot would have to keep polling. t3listen subscribes to t3 code's live event stream and pings the bot as soon as something happens. > permission prompts. even in full access, some runs still stop on "allow this command?" when a safety check flags it. mcp isn't built to handle those from outside, so while i'm afk tots can decline the risky ones by design and tell the agent to take a safer route, so the run keeps going simply put, it's more proactive, instantaneous, and doesn't burn your limits on polling while giving more functionality than what would be possible with current mcp but theo and team are overhauling t3 code's mcp, and once it's out i'll move tots onto it so it's more reliable and safer
This post quotes another post: “T3 Tots: a Grok Bot companion for T3 Code”