Using ChatGPT? Explore the Userorbit ChatGPT plugin for connection steps and example prompts.

MCP (Model Context Protocol) is the open standard that lets AI assistants call external tools. Instead of copying data between your AI tool and your product dashboard, an MCP server gives the AI direct access to the APIs it needs.

The Userorbit MCP server covers your whole workspace. Feedback, support inbox, announcements, roadmaps, help center, product tours, interactive demos, checklists, surveys, workflows, subscribers, and product analytics. It works with any MCP-compatible client: Claude, ChatGPT, Cursor, Windsurf, VS Code with GitHub Copilot, Claude Code, and others.

It's a hosted server. Paste one URL, sign in, and you're connected.

What you can do with it

AreaWhat you can do
Support inboxList and read threads with full timelines and customer context. Send customer replies, add internal notes, assign, label, merge, and mark as read. Run email or Slack side discussions. Manage labels and canned snippets.
FeedbackCreate, list, update, merge, archive, restore, and vote on feedback. Organize with boards and tags. Add public or private comments. See voters and subscribers.
AnnouncementsDraft, publish, schedule, pin, search, and archive changelog entries. Organize into collections. Pull completed feedback as release note evidence.
RoadmapsCreate roadmap views and columns. Read items grouped by status. Move items by updating the feedback status.
Help centerWrite articles, assign to collections, publish, unpublish, and archive. Full revision history with restore. Track helpfulness votes.
OnboardingManage product tours and tour steps, interactive demos, checklists, and surveys. Read tour progress, checklist progress, and survey responses. Validate saved definitions before publishing.
WorkflowsCreate, update, publish, and run support automation workflows. Inspect runs and diagnose failures.
AnalyticsQuery active and new users, look up a single person's events, and render charts inline in the chat.
ContactsCreate, search, count, update, and deactivate subscribers. Manage custom contact properties.
WorkspaceManage tags and projects. Search and fetch across feedback, announcements, articles, subscribers, and support threads.

Most tools follow the same pattern: pass an action (like create, list, or update) and a params object. Your AI picks the tool and action based on what you ask.

This embedded content uses optional tools. Enable analytics and embedded content to load it.

Setup

The server lives at one URL:

text
https://api.userorbit.com/mcp

There are two ways to authenticate.

  • OAuth. Add the URL as a custom connector and approve access in the browser. You pick the workspace and the scopes (read, write) on the consent screen. Manage or revoke connections under Settings → Connected Apps.
  • API key. Send your key as a bearer token. Get it from Settings → API in your workspace. Good for Cursor, CI jobs, and anything that can't run a browser sign-in.

Claude and ChatGPT

Open connector settings, add a custom connector, and paste https://api.userorbit.com/mcp. Complete the sign-in. The Userorbit tools show up in your next chat.

Claude Code

text
claude mcp add --transport http userorbit https://api.userorbit.com/mcp

Run /mcp inside Claude Code to finish the OAuth sign-in.

Cursor

Add to .cursor/mcp.json in your project root:

json
{
  "mcpServers": {
    "userorbit": {
      "url": "https://api.userorbit.com/mcp",
      "headers": {
        "Authorization": "Bearer your-api-key"
      }
    }
  }
}

Windsurf

Add to ~/.codeium/windsurf/mcp_config.json:

json
{
  "mcpServers": {
    "userorbit": {
      "serverUrl": "https://api.userorbit.com/mcp",
      "headers": {
        "Authorization": "Bearer your-api-key"
      }
    }
  }
}

VS Code (GitHub Copilot)

Add to .vscode/mcp.json in your project root:

json
{
  "servers": {
    "userorbit": {
      "type": "http",
      "url": "https://api.userorbit.com/mcp",
      "headers": {
        "Authorization": "Bearer your-api-key"
      }
    }
  }
}
On a multi-team account, add an x-team-id header with your team ID. OAuth connections pick the workspace on the consent screen instead.

Local stdio

Prefer running the server on your machine? The @userorbit/mcp package still ships a stdio server:

json
{
  "mcpServers": {
    "userorbit": {
      "command": "npx",
      "args": ["@userorbit/mcp"],
      "env": {
        "USERORBIT_API_KEY": "your-api-key"
      }
    }
  }
}

The stdio server covers the core feedback, announcement, help center, subscriber, and tag tools. Support, onboarding, workflows, and analytics are only on the hosted server.

Use case: reply to support threads from your AI tool

Your support inbox is now reachable from the same chat where you write code or plan the roadmap. Start with the queue:

text
Show me unassigned support threads from the last two days.

The AI calls userorbit_support_threads with queue: "unassigned" and lists subjects, customers, and priority. Open one:

text
Open the thread from Priya about CSV export failing.
Check if there's a help article that covers this.

It reads the full timeline with action: "info", searches userorbit_articles, and comes back with what the customer tried and which article applies. Then reply:

text
Reply to Priya with the steps from that article and a link to it.
Then add an internal note that this is the third export
complaint this week, and label the thread "exports".

The reply goes out through userorbit_support_reply. The note goes through userorbit_support_note, which can only ever write private notes. The label is set with action: "labels" on the thread.

Customer-facing replies have guardrails:

  • Separate tools. Replies and internal notes are different tools. An AI can't accidentally send a note to a customer.
  • Write scope required. Read-only connections never see the reply tool.
  • Safe retries. Every reply carries an operation key. If the network drops, the AI retries with the same key and the customer gets one message, not two.
  • Delivery receipts. The tool returns delivery state, and userorbit_support_operation checks it later. Saved doesn't mean delivered, and the AI can tell the difference.

Pull in the team without leaving the thread

text
Start a Slack discussion in #billing on this thread asking
whether we can refund the duplicate charge.

userorbit_support_discussions opens an internal email or Slack side conversation attached to the thread. The customer never sees it. Resolve it once you have the answer.

Turn repeat questions into docs

text
Look at resolved threads from this month. Find the questions
that came up more than three times and draft a help article
for the most common one. Show me the draft first.

The AI reads threads with queue: "all", groups the questions, and drafts the article in chat. Saving it to the help center is a separate step you approve.

Use case: automated changelogs from your AI editor

You finished a feature in Cursor. The code is committed. Instead of opening the dashboard to write a changelog entry, ask Cursor directly:

text
Draft an announcement titled "CSV export for feedback reports"
in the Product Updates collection. Mention that users can now
export filtered feedback as CSV from the feedback board.

Cursor calls userorbit_collections to find "Product Updates," then userorbit_announcements with action: "create". Add "publish it" or "schedule for next Monday at 9am" and the server passes publish or scheduledAt through to the API.

Release notes from shipped requests

text
Find feedback completed in the last two weeks and draft
release notes from it.

userorbit_completed_feedback returns requests marked completed, with pagination. The AI drafts the notes from real customer requests instead of commit messages. You review the draft in chat before anything is saved.

Use case: feedback triage in Claude

Product managers can triage feedback without opening the dashboard:

text
Show me all feedback items that are still in review.

Claude calls userorbit_feedbacks with action: "list" and returns titles, statuses, and vote counts. Follow up:

text
Mark the top three by votes as planned. Add a public comment
on each saying we're evaluating these for Q3. Merge the two
duplicate dark mode requests.

Claude updates statuses with userorbit_feedbacks, adds comments with userorbit_feedback_comments, and merges duplicates with action: "merge".

Organizing feedback with boards and tags

text
Create a feedback board called "Mobile App" and add the
"iOS" and "Android" tags to it.

The server creates the board through userorbit_feedback_boards, then associates tags through userorbit_feedback_board_tags.

Use case: roadmap updates from VS Code

Roadmap items are feedback entries. Roadmap views group them into columns by status. So moving an item means changing its status.

You're reviewing a PR in VS Code with Copilot. The feature is on the roadmap:

text
Mark "Webhook retry logic" as completed so it moves to
Shipped on the public roadmap.

Copilot finds the item through userorbit_roadmap_views with action: "items", then updates its status with userorbit_feedbacks. The roadmap stays accurate because the update happens where you have the most context.

Building a roadmap view

text
Create a roadmap view called "Q3 Platform" with columns:
Exploring (in review), Building (planned and in progress),
Shipped (completed).

The AI creates the view with userorbit_roadmap_views and adds each column with userorbit_roadmap_view_columns, mapping statuses to columns.

Use case: onboarding that keeps up with releases

Every release can break a product tour. A renamed button, a moved menu, a new step. The MCP server lets the AI check.

text
We renamed "Projects" to "Workspaces" in the sidebar.
Find tours and checklists that mention Projects and show
me the steps that need new copy.

The AI reads userorbit_tours, userorbit_tour_steps, and userorbit_checklists, then lists the affected steps with suggested copy. Apply the changes once you agree.

It works on the response side too:

text
How far do people get in the onboarding tour?
Summarize this week's NPS survey responses.

userorbit_tour_progress and userorbit_survey_responses return participant-level data. Before you publish a tour, checklist, or survey, the validate tools run static checks on the saved definition and list what failed.

Interactive demos

Demo screens are captured with the Userorbit Chrome extension. The MCP server handles everything after capture: edit step copy, duplicate, publish, read analytics, and generate a help article from a demo with action: "generateArticle".

Use case: product analytics in chat

text
Show me DAU, WAU, and MAU for the last 90 days.

userorbit_analytics_chart renders an interactive chart right in the conversation in Claude and ChatGPT. Each series gets its own interval, so this is one tool call, not three.

For raw numbers, userorbit_analytics_query returns the full series. userorbit_analytics_catalog lists exact event names so the AI doesn't guess them. To check on one customer before replying to their ticket:

text
What has priya@acme.com done in the product this week?

userorbit_analytics_person finds them by name, email, or ID and returns totals, per-event counts, and recent events grouped by session.

Use case: support workflows

Workflows automate the support inbox. The MCP server can build and debug them.

text
List our workflows. Why did the "Escalate enterprise
tickets" workflow fail yesterday?

The AI reads runs with userorbit_workflows and action: "runs", then pulls bounded diagnostics with userorbit_workflow_run_diagnostics. To change a draft, userorbit_workflow_draft_patch edits the step graph and validates it before you publish.

Use case: subscriber management

text
How many subscribers do we have?
Search for subscribers with "acme" in their info.

The AI calls userorbit_subscribers with action: "count" and action: "search". Custom fields like plan or company size live in userorbit_contact_properties.

Chaining operations

The leverage comes from combining tools in one conversation. A post-deploy routine from any MCP client:

text
We just shipped the billing dashboard redesign.

1. Draft a changelog entry in Product Updates
2. Mark the "Billing redesign" request as completed
3. Find open support threads about the billing page,
   reply with a link to the changelog, and close them
4. Check which tours mention the old billing page
5. Draft a help article about the new dashboard

That's announcements, feedback, support, tours, and articles in one prompt. What used to take an afternoon across five tabs happens in one conversation.

Weekly product review

text
Give me a summary:
- New feedback this week, top five by votes
- Support threads still unassigned
- WAU compared to last week
- Roadmap items in progress
- Unpublished draft announcements

One prompt pulls from five parts of the workspace.

Permissions and safety

  • Scopes. OAuth connections pick read or write. Read-only connections never see tools that change data.
  • Grant-filtered tools. The server only exposes tools your connection is allowed to call. userorbit_context shows the connected workspace and scopes.
  • Drafts first. Tool descriptions tell the AI to show drafts in chat before saving, and publishing is a separate step.
  • Conflict checks. Announcement and workflow edits carry a revision number. If someone else changed it in the meantime, the edit fails instead of overwriting their work.
OAuth grants store the list of allowed endpoints when you connect. If new tools appear and your connector doesn't show them, disconnect and reconnect it. API key connections pick up new tools automatically.

MCP vs. the Userorbit coding agent skill

If you've seen the Userorbit skill for coding agents, you might wonder which to use:

MCP ServerCoding Agent Skill
Works withAny MCP client (Claude, ChatGPT, Cursor, Windsurf, VS Code, Claude Code)Claude Code, Codex
SetupPaste a URL, sign in with OAuth or an API keyClone repo to skills directory
ProtocolModel Context Protocol (standardized)Agent-specific skill format
Best forTeams using multiple AI toolsTeams centered on a single coding agent

Both connect to the same Userorbit API. Pick MCP if your support team lives in ChatGPT, your PMs in Claude, and your engineers in Cursor. Everyone works against the same workspace.

Userorbit

Give your AI tools access to your product workspace

The Userorbit MCP server connects Claude, ChatGPT, Cursor, Windsurf, VS Code, and Claude Code to your support inbox, feedback, changelogs, roadmaps, help center, and analytics. One URL, zero code.

  • Reply to customer support threads from any MCP client
  • Triage feedback and keep roadmaps current from your editor
  • Draft changelogs from completed customer requests
  • Query product analytics and render charts in chat

Get started

Add https://api.userorbit.com/mcp to your AI tool, sign in, and ask it what's waiting in your support inbox.