TL;DR: You do not leave ChatGPT to use Dock. You connect Dock as an app or MCP server inside ChatGPT, and from then on your ChatGPT agent reads and writes Dock docs, tables, and files directly. The conversation still happens in ChatGPT. The output lands in a shared workspace your team and other agents can see, attributed and audited, instead of disappearing into the chat scroll.
ChatGPT is where a lot of work now starts. You shape a launch plan, draft a hiring rubric, or reason through a P&L, and the thinking is good. The problem is where it goes afterward. An hour later that brief is three threads deep in your sidebar, and a week later you are re-prompting to reconstruct a table you already built. ChatGPT is one-to-one and ephemeral by design. Dock is the persistent surface that fixes the ending without changing the start.
Can I use Dock without leaving ChatGPT?
Yes. Dock connects to ChatGPT over the Model Context Protocol, the open standard OpenAI uses to extend ChatGPT with external tools and data. Once connected, Dock is one of the tools your ChatGPT agent can call. You keep typing in ChatGPT, and the agent writes the result into a Dock workspace in the same turn.
This is the core idea of an MCP-first workspace: the surface an agent works on should be reachable from inside whatever client the agent already lives in. You are not asked to adopt a new chat window. ChatGPT stays your front end. Dock becomes the back end the conversation persists to.
What does connecting Dock to ChatGPT actually do?
It turns ephemeral conversation into durable artifacts. Instead of a wall of text you have to copy-paste out later, your ChatGPT agent creates a Dock doc, drops rows into a Dock table, or saves a file, and that object lives in a workspace with a stable URL.
Three things become possible the moment the work lands in Dock rather than the chat. First, persistence: the doc is still there next week, searchable and editable, not pruned with your history. Second, attribution: every write is signed, so you can see which agent and which person produced it. Third, review: a teammate, or another agent, can open the same workspace and build on it. This is what an AI workspace is for, a place where agent output is a first-class object instead of a transcript you have to mine.
Dock plus ChatGPT, not Dock versus ChatGPT
ChatGPT is a partner here, not a competitor. The two do different jobs and the combination is stronger than either alone. ChatGPT is the reasoning surface and the conversational interface. Dock is the system of record the reasoning writes to. You bring your own ChatGPT subscription, your own prompts, your own habits. Dock just gives the output somewhere to live.
This matters because the alternative is worse for everyone. Without a shared workspace, every good ChatGPT session ends with manual copy-paste, lost context, and work that only one person can see. With Dock connected, the session ends with a durable, shared, attributed artifact, and ChatGPT did not have to change at all. The platform you already use gets a memory and a team layer it did not have before.
How to set this up
There are two ways to connect Dock to ChatGPT, covered in depth in Dock inside ChatGPT. Both speak the same backend, so you pick whichever fits how you work.
The first is the curated Dock app on OpenAI's Apps SDK. It exposes a focused set of tools with inline widgets, a workspace picker, a doc preview, and an Approve or Cancel card, so common actions feel native inside ChatGPT. The second is the full Dock MCP server, which exposes every Dock primitive for power users who want the complete tool surface. OpenAI documents both connection paths in its own MCP developer guide.
Setup is short:
- Open ChatGPT Settings, then the Apps or connectors section.
- Add Dock, either the curated app or the MCP server endpoint.
- Complete the OAuth handshake so ChatGPT connects to your Dock account.
- Pick a workspace, then ask ChatGPT to write to it.
After that, prompts like "save this plan as a Dock doc in the Launch workspace" or "add these candidates as rows in my hiring table" do exactly what they say.
How Dock approaches agent identity and safety
Connecting an agent to a real workspace raises a fair question: who is acting, and what are they allowed to do? Dock's answer is that an agent is a principal, not a costume a human wears. Your ChatGPT agent connects through the OAuth handshake and acts under a signed agent identity with its own scopes, rather than silently borrowing your credentials.
That has practical consequences. Every write your ChatGPT agent makes to Dock is attributed and recorded, which gives you a real audit trail across human and agent activity instead of an anonymous edit. Scopes mean the agent can do exactly what you granted and nothing more. And for irreversible or sensitive actions, Dock puts a consent gate in front of the operation, so a destructive call needs explicit approval before it runs. The result is a workspace you can hand to an agent without giving up the ability to see and govern what it did.
If you are weighing options across platforms, the same identity and audit model is why Dock holds up as a workspace built for AI agents, not a human tool with a chatbot bolted on.
Connect Dock to ChatGPT
If your team already runs on ChatGPT, the fastest upgrade is not a different model. It is giving the model a place to put its work. Connect Dock inside ChatGPT, point your agent at a workspace, and let the next good conversation end as a durable artifact instead of a lost thread. See the full ChatGPT integration and install steps.
FAQ
Do I need to stop using ChatGPT to use Dock?
No. That is the whole point of the integration. You keep ChatGPT as your conversational front end and connect Dock as a tool inside it over MCP. Your ChatGPT agent reads and writes Dock surfaces directly, so nothing about your ChatGPT workflow changes except where the output lands.
How does ChatGPT connect to Dock technically?
Through the Model Context Protocol, the open standard OpenAI supports for extending ChatGPT with external tools and data. Dock offers a curated app on OpenAI's Apps SDK and a full MCP server, both reachable from ChatGPT's Apps or connectors settings. You complete an OAuth handshake once, then your agent can call Dock tools in any conversation.
Who gets credit for the work my ChatGPT agent does in Dock?
Every write is attributed to a signed agent identity tied to your account, so the record shows which agent acted and on whose behalf. This is different from an agent borrowing your login and editing anonymously. It gives you a clean audit trail and lets teammates trust what shows up in a shared workspace.
Is it safe to let a ChatGPT agent write to my workspace?
Yes, because the agent acts within granted scopes rather than with full account access, and sensitive or irreversible operations sit behind a consent gate that requires explicit approval. Combined with full attribution and an audit log, that means you can see exactly what the agent did and stop anything it should not do. Safety is built into the identity model, not added afterward.