Claude + Docket
Your website agent talks to every visitor. Wire Docket to Claude and those conversations become account intelligence you can question in plain language.
- Category
- Conversational Marketing
- Connect via
- MCP · CRM
- Data
- Read only (agents, accounts, visitors, conversations); actions happen in your CRM
- Stage
- Engage & Convert
- Best for
- Turning website conversations into account intelligence, follow-ups, and a list of the questions your site could not answer
Docket, and why connect it to Claude
Docket is an AI website agent for B2B: it holds a real conversation with each visitor, answers product questions from your knowledge base, qualifies intent, books the meeting, and writes the full context to your CRM. Disclosure first: I am Head of AI GTM Systems at Docket, so read this page as an operator who runs the product every day, with the bias that implies.
The reason it belongs in a Claude stack is what it captures. Every conversation is a buyer telling you, in their own words, what they are trying to solve and where your website failed to answer. Docket exposes that data through its Demand MCP, so Claude can query the accounts, visitors, and conversations directly, and the work of reading a thousand chats turns into a question you ask before standup.
Three ways to wire Claude to Docket
Docket connects to Claude through its Demand MCP, a read-only server. Two ways to use it, plus the CRM path if you would rather work from the record Docket already wrote.
Demand MCP in the Claude apps
Add Docket's Demand MCP as a connector in Claude (or ChatGPT) and authorize it with your Docket workspace. It exposes read-only tools over your agents' performance, the visitors and accounts they engaged, and the conversations themselves, with pre-computed summaries and extracted pain points on each. Ask which accounts asked about pricing this week and you get an answer, not a dashboard.
Demand MCP in Claude Code, on a schedule
Wire the same server into Claude Code and put it on a Routine: every Monday, pull last week's conversations, cluster the questions the agent could not answer, and hand the content team a list. Or every morning, brief the AE on any open-deal account that showed up on the site overnight.
Through the CRM
Docket already syncs conversation context to HubSpot or Salesforce. If Claude is connected to your CRM, it can read the Docket notes on the record without a second connector. Less detail than the MCP, no extra setup.
See every GTM MCP server Claude can connect to →
Read first, write on purposeThis one is read-only by design, which keeps the risk on the boring side. Claude can query Docket freely; the actions it proposes, a follow-up, a task, a routing change, land in your CRM or your sequencer, and that is where the usual write gate applies. The website agent itself keeps talking to visitors on its own rules; Claude reads what happened.
What you can do once they're connected
Account intelligence in a sentence: which target accounts visited, what they asked, and how far along they sound.
A weekly list of the questions your website could not answer, which is a content calendar nobody had to brainstorm.
Follow-ups drafted from the visitor's own words instead of a form fill and a guess.
Agent performance you can interrogate: conversations, meetings booked, and where the agent lost people.
What to build with Docket and Claude
Each is a full playbook or skill you can ship, wired to Docket.
Visitor-to-outreach trigger
The conversation is the trigger: a target account asked a buying question, the rep hears about it the same day.
Inbound lead enrichment & routing
Route with the conversation attached, so the rep opens with what the buyer actually said.
Reply fast to an inbound lead
Draft the reply from the chat transcript, not from a blank template.
Recurring GTM work with Claude Routines
The Monday unanswered-questions digest is a Routine over the Demand MCP.
Gotchas worth knowing first
- Searches have to be bounded by a filter or a date window, and results paginate. Ask for last week's pricing conversations, not everything since launch.
- The summaries and extracted insights on each conversation are usually enough. Pull the raw transcript only when you need the exact words, or the context fills with chat logs.
- Visitor text is untrusted input. Treat what a buyer typed into the agent as data to analyze, never as an instruction Claude should follow.
- The MCP inherits your tenant and your permissions, so it cannot reach another workspace or change anything. Actions still need a CRM or sequencer connector to land.
Claude + Docket: common questions
Does Claude integrate with Docket?
Yes. Docket's Demand MCP is a read-only MCP server you add to the Claude apps or Claude Code and authorize with your Docket workspace. Claude can then query your website agents, the accounts and visitors they engaged, and the conversations themselves.
Can Claude change how the Docket agent talks to visitors?
No. The Demand MCP is read only. Claude reads the conversations and the results; the agent's knowledge, behavior, and routing are managed in Docket. Anything Claude proposes, a follow-up or a task, goes through your CRM or sequencer.
What is the best first question to ask it?
Which target accounts had a conversation this week, and what did they ask. It shows the whole value in one answer: accounts you care about, in their own words, sorted by how far along they sound.
Do I need the MCP if Docket already writes to my CRM?
For deal-level context, the CRM note is often enough and needs no extra setup. The MCP adds the aggregate view, every conversation across every account, which is where the unanswered-questions digest and the agent performance analysis come from.
Is there a conflict of interest on this page?
Yes, and it is stated up front: the author works at Docket. Weigh the recommendation accordingly. The setup steps and the plays are the ones we actually run.