TL;DR: You do not leave Claude to use Dock. Claude supports custom connectors over the Model Context Protocol, and Dock exposes its full workspace as a remote MCP server. Connect the two and your Claude agent reads and writes Dock docs and tables from inside Claude, so the research, drafts, and decisions a chat produces land in a persistent, attributed, shared record instead of evaporating with the thread.
How do I use Dock inside Claude?
You connect Dock to Claude as a custom connector. Claude adds remote MCP servers as connectors, and Dock publishes its workspace as one of those servers. Once connected, Claude can call Dock's tools mid-conversation: create a workspace, append a section to a doc, drop rows into a table, search across everything you have.
The result is that Claude becomes a member of your workspace, not a separate tool you copy-paste out of. You talk to Claude, Claude writes to Dock, and the work stays where your team and your other agents can find it. This is the model we describe in what an AI workspace actually is: a shared place that humans and agents both write to, rather than a chat that only you can see.
Does Claude support MCP and custom connectors?
Yes. Claude supports custom connectors built on remote MCP servers across Claude, Cowork, and Claude Desktop, for users on Free, Pro, Max, Team, and Enterprise plans, per Anthropic's connector documentation. Free accounts are limited to a single custom connector; Pro and Max users add connectors directly; Team and Enterprise owners add a connector for the organization and members connect to it individually.
Two accuracy notes worth keeping straight. Remote connectors reach your server from Anthropic's cloud infrastructure, so the MCP server has to be reachable over the public internet, which Dock's hosted server is. Separately, Claude Desktop also supports local MCP servers through desktop extensions, a different path for tools that run on your own machine. Dock's connector is the remote kind: one URL, one OAuth sign-in.
Why connect Claude to Dock instead of working in chat alone?
Because a chat thread is a single-player, forgetful surface, and most real work is neither. A Claude conversation is excellent for thinking. It is a poor place to keep the output, because the thread is yours alone and the context ends when the thread does.
Dock is the durable other half. When Claude writes a brief to a Dock doc or a set of leads to a Dock table, that artifact persists, your teammates can open it, and your next Claude session can read it back. You get the reasoning power of Claude and the memory and shared structure of a real workspace, which is the whole argument for building an MCP-first workspace rather than bolting agent access onto a human-only tool.
Claude + Dock: which surface does what
Think of the two as complementary halves of one loop, not competitors.
- Claude is the reasoning surface. It drafts, analyzes, summarizes, and decides. It is where the human conversation happens.
- Dock is the system of record. It holds the persistent docs and tables that the conversation produces, attributed to whoever or whatever wrote them.
The handoff is the point. A chat in Claude that produces a launch plan is valuable for an hour. The same plan written to a Dock workspace, where the punch-list table updates as work lands and the positioning doc stays current, is valuable for the life of the project. Claude reasons, Dock keeps. You move between them without ever leaving the Claude window, because Claude is calling Dock's tools for you.
This is also why Dock works the same way no matter which model you run. The same workspace your Claude agent writes to is the one ChatGPT can write to over MCP. The workspace is model-agnostic by design, so you are never locked to one assistant to read your own record.
How Dock approaches this
Dock is an MCP-native workspace, which means the agent-facing path is the primary one, not a courtesy layer. A few mechanisms make the Claude connection trustworthy rather than just convenient.
Signed agent identity. When your Claude agent writes to Dock, it does so under an identity, not a borrowed human login. Every doc edit and row insert is attributed to a specific principal you can see and revoke. We make the full case for this in why agents need their own identities: an agent that acts as itself is one you can actually govern, and agent identity is the foundation the rest of the model rests on.
OAuth, per-user, scoped. You connect Claude to Dock with a single OAuth sign-in. The agent inherits your workspace permissions: it sees only workspaces you belong to and writes only where you have editor access. Remove yourself from a workspace and the agent's access to it goes with you.
Dual-keyed audit and consent gates. Every action lands in an append-only record, so you can answer who changed what and when after the fact. This is the backbone of agent audit and compliance. Destructive operations, deleting a workspace or changing member roles, route through a two-call confirmation: the agent's first call returns a token plus a plain-language summary, you approve, and only then does it commit. Claude surfaces that summary in the conversation before anything irreversible happens.
The workspace is the canonical record. Dock is the system of record for agent output. Claude's MCP calls read from and write to that same canonical store every other client sees, which is why this approach holds up as you add more agents. It is the reason Dock reads as the best AI workspace for AI agents rather than another chat tool with an API stapled on.
How to set it up
The flow is short. In Claude, open your connector settings and add a custom connector. Paste Dock's MCP server URL and choose OAuth as the authentication method. Claude opens a browser tab, you sign into Dock and grant the agent access, and the connection is live. From then on, ask Claude to pull up a workspace, save the current draft to a doc, or add rows to a table, and it calls the matching Dock tool, surfacing a confirmation for anything destructive before it runs.
Connect Claude to Dock and give your agent a workspace its work survives in. Start with Dock and add the connector in a couple of minutes.
FAQ
Do I have to leave Claude to use Dock? No. That is the entire point of the connector. Once Dock is added as a custom connector, your Claude agent reads and writes Dock surfaces from inside the Claude conversation you are already in. You stay in Claude; the durable artifacts land in Dock.
Does Claude actually support MCP and custom connectors? Yes. Per Anthropic's documentation, custom connectors via remote MCP servers are available on Claude, Cowork, and Claude Desktop across Free, Pro, Max, Team, and Enterprise plans. Free accounts get one custom connector; paid plans add more, and Team and Enterprise owners can provision a connector for the whole organization.
Is Dock a replacement for Claude? No, they are complementary. Claude is the reasoning and conversation surface; Dock is the persistent, shared workspace that the conversation writes into. You use them together: Claude drafts and decides, Dock keeps the record. Neither one does the other's job well alone.
Who gets credit when Claude writes to Dock? The work is attributed to a signed agent identity acting under your OAuth grant, recorded in Dock's append-only audit trail. You can see exactly which agent made each edit, the agent inherits only your permissions, and destructive actions require a separate confirmation step before they commit.
