How to use Docket to Analyze Website Demand Data in LLMs With Demand MCP


Docket Demand MCP connects your Docket data to ChatGPT, Claude or any MCP client. You ask a question in plain language, like which target accounts talked to your agent this week, and get an answer drawn from every conversation, visitor and account your Demand Agent captured, with conversation IDs to check it.
Your website agent hears what buyers need in their own words: the pain point, the competitor they're comparing, the question your site didn't answer. Most of that sits in a dashboard until someone has time to filter it, export it or file a report request. By then the account has gone quiet.
Docket Demand MCP is an MCP server that gives your AI assistant governed access to your Docket demand data. MCP (Model Context Protocol) is the open standard AI apps like Claude and ChatGPT use to connect to outside tools and data. Docket is The Demand Platform, and Demand MCP is how its data reaches the assistant your team already works in.
It works in Claude, ChatGPT, Claude Code, Codex and any MCP client that supports sign-in with OAuth. Docket's app is also live in ChatGPT.
You can ask Demand MCP about your agents, visitors, accounts, conversations and meetings, in plain language. Each conversation comes pre-analyzed, so the assistant reads summaries and extracted insights instead of opening transcripts one by one.
| Ask about | Example question | What comes back |
|---|---|---|
| Agent performance | "How did our agents do this month versus last?" | Visits, conversations, engagement rate, leads, meetings booked and CTA clicks, with period-over-period change |
| Accounts | "Which of our 20 target accounts visited in the last 30 days?" | Accounts with enrichment, lead counts and the visitors behind them |
| Visitors | "Who came back three times this week, and from which campaign?" | Profile, enrichment, UTM and referrer journey, consent state |
| Conversations | "What objections did buyers raise in October?" | Summary, pain points, interested products, discovery questions and next steps per conversation |
| Meetings | "Which booked meetings were rescheduled or cancelled?" | Every booking the agent made, by status |
| Knowledge gaps (V2) | "Which questions did the Pricing agent fail to answer last week? Fix them." | The unanswered questions, then the knowledge to add or prompt change to make, published when you say so |
| Custom signals (V2) | "Which competitors did buyers mention this month?" | Answers from a post-call field you set up in chat, also written back to your CRM |
| Integrations (V2) | "Which integrations are connected, and is anything unhealthy?" | Every integration with its setup and health status, plus a sign-in link for anything not connected |
| Exports (V2) | "Export last month's conversations and meeting bookings." | Files of visitors, accounts, conversations, bookings, daily agent performance or transcripts |
Marketing teams use Demand MCP to turn buyer conversations into lists, content and reports without waiting on an export. The most common jobs:
Demand MCP V2 covers a Website Agent's whole lifecycle from the chat: build it, tune it, connect it to your stack, measure it and take the data out. Spot a weak answer or a missed booking, then fix it without opening the dashboard: "Read last week's conversations for the Pricing agent, find the questions it couldn't answer, and fix them."
| Area | What admins can do from the chat |
|---|---|
| Create and clone | Start a draft agent from a starter template, or clone an existing agent and choose what to leave out |
| Prompt | Describe a change in plain language, answer any clarifying questions, review the draft and publish when you say so |
| Persona and handoff | Set voice or text mode, persona name, greeting and avatar, plus human handoff instructions and Slack or Teams channels |
| Meetings and CTAs | Set booking triggers, CRM or agentic routing, calendar, meeting length and reminders, and add, edit or turn off CTAs |
| Knowledge | Add a URL, sitemap, file or pasted text, choose the sources, slides and videos the agent uses, and resync pages |
| Post-call analysis | Add or edit the fields your agent extracts from every conversation |
| Agent settings | Set CRM writeback and field mappings, FAQs, consent, working hours, page targeting, IP rules and post-call alerts |
| Integrations | See every integration's setup and health, get the sign-in link in chat, finish setup after a one-time browser sign-in, and disconnect with a keep-data or delete-data choice |
| Exports | Export visitors, accounts, conversations, meeting bookings, daily agent performance and transcripts as files you can load anywhere |
| Deployment | Change widget text, look and position, manage allowed domains, and get the embed script, standalone page link or iframe code |
Every change needs admin rights. Publishing a prompt waits for your explicit instruction, and archiving an agent shows a preview first and asks you to type the agent's name.
To turn on V2 for your account, talk to your Docket customer success manager.
A dashboard answers the questions someone set it up to answer. Demand MCP answers the question you have right now, and it can combine Docket data with other connectors in the same chat.
| Dashboard | CSV export | Demand MCP | |
|---|---|---|---|
| How you get an answer | Filter and click | Download, then build it in a spreadsheet | Ask in plain language |
| Questions it handles | The ones it was built for | Anything, after manual work | Anything your data can answer |
| Works with your other tools | No | Only by hand | Yes, alongside connectors like your CRM in the same chat |
| Proof behind the answer | Charts | Raw rows | Conversation IDs you can open |
Demand MCP is scoped to your company's data and to each person's role. You sign in with your Docket work account, every request is isolated to your tenant and logged, and roles set how deep each person can go. Everyone with access can read performance. Transcripts and call recordings stay with admins and contributors, and links to them expire after 15 minutes.
You connect Demand MCP by adding it as a connector in your AI app and signing in with your Docket work email. In Claude, go to Settings, then Connectors, add a custom connector with Docket's Demand MCP URL, and sign in. In ChatGPT, add the Docket app. No developer is needed, and your first question can go in as soon as you've signed in.
Demand MCP works with Claude, ChatGPT, Claude Code, Codex and any MCP client that supports OAuth sign-in. Your team can use whichever assistant they already have, and each person sees the data their Docket role allows.
No. Demand MCP is for your team: it brings your Docket data into your own AI assistant. Docket WebMCP is for your buyers: it lets a visitor's AI assistant ask your website questions and hand off to your Demand Agent.
Yes, if your AI app has those connectors too. Demand MCP sits next to them in the same chat, so one prompt can check Docket conversations against open opportunities and owners in your CRM.
Admins and contributors can open transcripts and call recordings, and those links expire after 15 minutes. Other users can read performance, accounts and conversation analysis without the raw transcript.
No. Every conversation comes pre-analyzed with a summary, pain points, interested products, discovery questions and next steps. The assistant works from that analysis and opens a transcript only when you ask for the exact words.
See how it works at docket.io/features/demand-mcp, or ask Aura on docket.io.