Pick up where you left off, in any AI tool.

Teb carries your working context between Claude, ChatGPT and any tool that speaks MCP: the goal, what was decided, where you stopped and what is still open. One line to save, one line to continue.

Get started free How to install10 handoffs a month free.

By signing in or connecting Teb you agree to the Terms and Privacy Policy.

When you need it

ChatGPT is down.

Say “Teb T-4K2P” in Claude and keep going.

Research in one, code in another.

Finish the research in Gemini, write the code in Claude with everything decided.

Out of context window.

Save a clean summary, start fresh anywhere, nothing already decided gets re-asked.

How it works

In any tool: “save to Teb” (in ChatGPT: “@Teb save this chat”). Your AI summarizes the goal, the decisions, where you stopped and what is still open. You get a code like T-4K2P.
In the other tool: “Teb T-4K2P” (in ChatGPT: “@Teb continue T-4K2P”). It reads the handoff and continues from where you stopped, without reopening what was decided.
No Teb there? Open the share link and paste the prompt. Works in every AI tool.

What Teb carries

Goal and decisions

What you are building and what is already settled, marked final so the next tool does not reopen it.

Where you stopped and what is open

The exact next step, plus the questions still waiting, so work resumes instead of restarting.

Your style and materials

Language, tone and format preferences, with the code, links and drafts the work depends on.

Why not copy and paste? Pasting gives the next AI raw text to re-read. Teb gives it a structured working state written as an instruction. And for a project that moves between tools many times, save into a thread: every handoff becomes a version, the latest state is one line away, and decisions you changed along the way stay in the history as superseded instead of disappearing.

Install

Teb is a remote MCP server. Add it once per tool, sign in, done.

https://tebhq.com/api/mcp

Claude

Settings → Connectors → Add custom connector → paste the URL → Add. Sign in with your Teb account.
Web, desktop and mobile. Claude Code: run claude mcp add --transport http teb https://tebhq.com/api/mcp.

ChatGPT

Settings → Plugins → Add → Create custom MCP server. Name Teb, paste the URL as Server URL, leave Authentication on OAuth, tick “I understand”, Create as a plugin → Continue to Teb. Sign in with your Teb account.
In a chat, type @Teb, pick it from the popup, then write “continue T-XXXX” or “save this chat”. Without the mention ChatGPT may answer from memory instead of calling Teb.

Cursor

Settings → Tools & MCP → Add custom MCP server. Add this to the JSON that opens, then click Connect next to teb:

{
  "mcpServers": {
    "teb": {
      "url": "https://tebhq.com/api/mcp"
    }
  }
}

Other MCP clients

Any client that supports remote servers over Streamable HTTP with OAuth: add the URL above and sign in when prompted. The same JSON as Cursor works for Windsurf, VS Code and Claude Code config files.

Gemini, Grok, anything without MCP

No install. Every handoff has a share link. Open it, copy the prompt, paste it into the tool.

Plans

Free

10 handoffs a month
Kept 30 days, full transcripts 7 days
Threads and search included
Resuming is always free

Tokens (coming after launch)

Unlimited handoffs, kept forever
Full thread history, never expires
Hand off to several tools at once

Privacy

Handoffs are tied to your account and only you can resume them. A handoff is the structured summary. For long conversations, or when you ask, your AI also attaches the full transcript and tells you so; say “without the transcript” to skip it. Transcripts are encrypted at rest, returned only when a tool asks for them, and deleted after 7 days on the free plan. The share link carries a secret key, so only someone holding the link can read a handoff on the web. Teb does not run a model on your data. Details in the privacy policy.

FAQ

Which tools? Any MCP client with remote connectors: Claude, ChatGPT, Cursor and more. For tools without MCP, the share link works everywhere.

Does the other tool really not re-ask? The handoff text is written as an instruction: what is decided is marked final, and the first step is to continue from where you stopped. The target tool works from that text, so a good handoff is a complete one.

Is it the same conversation, with the other model’s memory? No. AI tools do not expose their internal state, and Teb does not pretend otherwise. Teb carries a structured handoff: goal, decisions, where you stopped, open items, your style and materials, and for long conversations the full transcript as well. The other tool continues from that.

What is a thread? A project that lives across many handoffs. Say “save to Teb, thread Nightlife App” and later “Teb continue Nightlife App” in any tool: you get the latest state, which decision came from which handoff, and what was decided earlier and later changed.

Who summarizes? The AI you are already using, following Teb’s schema. What leaves your tool is that summary, plus the full transcript when the conversation is long or you ask for it. Your AI tells you when it attached one.

Can I see how it is built? The architecture, the controls and how to report a vulnerability are on the Security page.