The Appcues MCP Server connects AI tools like Claude, ChatGPT, and Cursor to your Appcues account so you can query experiences, analyze data, and manage workflows using natural language instead of navigating the dashboard.

What the MCP Server does

MCP (Model Context Protocol) is an open standard that lets AI assistants interact with external tools and data. The Appcues MCP Server uses this protocol to give your AI client secure access to your Appcues account — Flows, experiences, segments, goals, NPS data, events, and more. Access is read-only, with one exception: your AI client can publish and unpublish experiences. The server does not expose tools that edit or delete your content.

You work with your AI client using natural-language requests. That client uses the MCP Server to make the right Appcues tool calls, fetch the data, and return it in a conversational format.

Prerequisites

• An active Appcues account with admin access (to enable the MCP Server)
• An MCP-compatible AI client (Claude, ChatGPT, Cursor, or similar)

Set up the MCP Server

Open Appcues Studio > Settings > Appcues AI > AI Assistant. You can go directly to studio.appcues.com/settings/ai-assistant.

Toggle the MCP Server setting to On. This enables MCP access for your account.

Open your AI client and register the Appcues MCP Server URL:

https://mcp.appcues.com/mcp

For accounts on Appcues' EU data hosting, register the Appcues MCP Server URL:

https://mcp.eu.appcues.com/mcp

Which URL should I use? Most Appcues accounts use the standard URL (mcp.appcues.com/mcp), regardless of where your company is located. The EU URL is only for accounts that have specifically opted in to EU data hosting. When in doubt, start with the standard URL.

Follow your client's documentation for adding an MCP server:

• Claude: Connect remote servers
• ChatGPT: Connectors in ChatGPT
• Cursor: MCP context

Connect to the Appcues MCP Server from your AI client. Enter your Appcues credentials when prompted.

Example Appcues MCP Setup with Anthropic Claude

Confirm it worked

Ask your AI client a simple question like "List my published flows." If it returns your actual Appcues data, you're connected.

Use cases

Here's where the MCP Server becomes practical. These are real workflows you can run through natural-language prompts in your AI client.

Pull experience performance summaries

Ask your AI client to summarize how a specific flow or set of flows is performing. Instead of clicking through analytics screens, get completion rates, skip rates, and unique user counts in one response.

Try: "What are the completion rates for my onboarding flows this month?"

Monitor your NPS score

Query your current NPS score, see the breakdown of promoters, passives, and detractors, and review recent open-text responses — all without opening the NPS dashboard.

Try: "What's my current NPS score, and what are the most recent responses?"

Audit your published experiences

Get a quick inventory of everything that's live. Useful for quarterly reviews, handoffs, or making sure nothing outdated is still running.

Try: "List all published experiences and when they were last updated."

Investigate segment membership

Check how many users are in a given segment, or explore what conditions define it. Helpful when you're building a new flow and want to confirm your targeting before you publish.

Try: "How many users are in my 'Trial users — day 7' segment?"

Track goal completion

See how many users have completed a specific goal, and pull the list of user IDs who hit it. Combine this with segment data to understand which cohorts are converting.

Try: "How many users completed the 'Activated' goal this quarter?"

Review survey responses

Pull survey answers from a specific experience without exporting a CSV. Useful for quick reads on qualitative feedback.

Try: "Show me the survey responses from my feature request flow."

Triage content issues

Ask for recent experience errors or content issues. This surfaces problems (broken steps, rendering issues) that might be affecting live users.

Try: "Are there any content issues with my published flows?"

Automate reporting across tools

If your AI client supports multiple MCP servers or tool integrations, combine Appcues data with other sources. For example, pull Appcues flow performance into a Slack summary, or cross-reference NPS scores with data from your CRM.

Try: "Summarize this week's Appcues NPS responses and post a summary to #product-feedback in Slack."

Security and permissions

Common questions from security and DevOps teams evaluating the MCP Server.

Does the MCP Server expose end-user personal data?

That depends on the data you send to Appcues.

• If you send only anonymous data — user UUIDs with no names or email addresses — no personal data can surface through the MCP Server, because Appcues never receives it.
• If you send end-user data, or you're not sure, you control whether AI tools can reach it with the Allow Appcues enhanced AI features that will send your Users data to Appcues' LLM Provider setting in Settings > Appcues AI > Enhanced AI features.

When that setting is off, your AI client can't use tools that return user-level data — for example, "list the users who saw this Flow." Aggregate analytics still work — for example, "how many users saw this Flow."

Can the MCP Server change or delete our data?

No. The server is read-only apart from publishing and unpublishing experiences. There are no tools that edit or delete content, so an AI workflow can't corrupt or remove your data.

What permissions does an MCP connection have?

The connection uses your own Appcues login, so it has exactly the permissions your role has in Studio — no more. A non-admin connecting through MCP can't perform admin actions.

On Claude Teams and Enterprise, admins can authorize which connectors are available, but each person connects with their own credentials. Admins don't share or hold those credentials.

Things to know

• Client-specific quirks exist. Claude currently supports remote MCP connections from desktop and web (not mobile). ChatGPT supports MCP from mobile and web (not desktop).
• Appcues connector must be enabled/toggled on. In ChatGPT or Claude, in any given chat and you need to click the + button in the chat to attach the Appcues connector before querying.
• Appcues Captain AI uses the same tools exposed by Appcues MCP Server. The difference is that Captain AI also leverages specialized agent rules and context and enhanced client support in Appcues Studio. 
• New tools are added regularly. If you need a capability that isn't available yet, reach out to support.

Troubleshooting

You're not getting Appcues data

• Verify the MCP Server is enabled in Settings > Appcues AI.
• Confirm you've attached the Appcues connector in your chat session (required for ChatGPT and Claude).
• Check that your Appcues role has permission for the data you're requesting.

The connector shows connected but requests fail

If the connector looks connected but requests fail, or it reports an expired or disconnected token, refresh your session:

1. Disable and re-enable the Appcues connector in your AI client. If that doesn't restore access, continue.
2. Log out of Appcues Studio completely, then log back in.
3. Reconnect the Appcues connector and re-enter your credentials.

This clears a stale login token, which is the most common cause of this state.

You get a connection error the first time you connect an account

Confirm the MCP Server is toggled On in Settings > Appcues AI for that specific account. When access hasn't been enabled in Studio yet, your AI client shows a generic error that doesn't name the real cause.

If it's still not working

Collect your Appcues account ID, the AI client you're using, the prompt you sent, and any error text, then contact Appcues support at support@appcues.com.