Notes from building Tokenrip, the shared workspace for your AI agents.
Claude Duet uses the connectors, memory and skills already in your Claude account. What Anthropic documents about Duet's connectors, memory, usage limits and the waitlist, how it compares with dots and Muse, and what we plan to test when our invitation arrives.
Grok Bot memory explained: what each Bot keeps across sessions, why memory leaks between projects on one Bot, what sharing a Bot or using Team Bots does to memory, how to reset it, custom MCP servers, and how we run Grok Bots as our daily agents.
Grok memory explained: what grok.com remembers across chats, why it can seem forgetful, how to turn it off, how it differs from ChatGPT memory and from Grok Bot and Grok Build memory, custom MCP connectors, and why we use Grok every day without relying on its memory.
Meta Muse has no settings form for MCP servers: you ask Muse to build a custom connector on its own cloud computer. How that works, what Meta reviews and does not, where Muse keeps keys and memory, whether two Muses can collaborate, and our plans for a Tokenrip connector.
ChatGPT dots memory explained: your dot starts from your ChatGPT memory and keeps its own notes you cannot view or edit, reset is the only wipe, what it shares with ChatGPT and Codex, whether it can use a custom MCP plugin, and what we found using one.
Claude Duet, OpenAI dots, Meta Muse and Grok Bot compared on plans, price, regions, memory control, custom MCP servers and team use, from the makers' pages, plus which ones we run at Tokenrip and why Grok Bot won.
How to add an MCP server to ChatGPT: the Plugins flow, Developer mode, the desktop app, which plans qualify, OAuth, local servers through Secure MCP Tunnel, and the fixes for the usual problems.
Claude and ChatGPT cannot read each other's chats, projects or memory. What works instead: copy and paste, a shared Google Drive or Notion, or one MCP workspace both read and write. Plus how we split the work between them every day.
Claude Code for teams in 2026: what a team shares through the repo (CLAUDE.md, rules, skills, MCP config, settings), what stays per person (memory, sessions, projects), the Team plan, agent teams mode, and how we run it as a two-founder company.
Claude Code handoff vs /compact, how to continue a session on another machine or your phone, a copyable handoff prompt and /handoff skill that also works in Codex, and how we hand off between agents without a HANDOFF.md.
