The Reprise MCP: Meet your AI Demo Engineer

The Reprise MCP (Model Context Protocol) connects your Reprise account to the AI assistant you already use — Claude, ChatGPT, Codex, Copilot, and others — so you can build, edit, and customize demos with natural-language prompts instead of clicking through the editor.

Point your AI assistant at a demo and tell it to "rebrand this healthcare demo for retail" or "swap the logo on every screen in my demo library." Your assistant does the work for you — it uses the Reprise MCP to read your demos, understand their structure, and apply the changes you asked for. What used to take hours now takes one prompt.

This article covers what you can do with the Reprise MCP, how to set it up, and the questions security and IT teams typically ask.

What you can do with the Reprise MCP

The Reprise MCP gives your AI assistant 32 tools that span Reprise's three product lines — HTML, Clone, and Data Injection. Anything you can do in the Reprise platform, you can ask your AI assistant to do for you.

Your AI assistant (Claude, ChatGPT, Copilot, etc.) is the one doing the work. The Reprise MCP is what lets that assistant read your Reprise content and apply changes back to it. The "intelligence" — understanding the prompt, deciding what to change, writing the new copy — comes from your assistant. The MCP is the bridge that lets the assistant act inside Reprise on your behalf.

A few examples of what customers are using it for today:

• Rebrand a demo for a new vertical or prospect. Point your assistant at a demo and ask it to "copy this demo and rewrite all the guides and on-screen text for a retail audience instead of healthcare." Your assistant uses the Reprise MCP to read every text node and guide, then writes and saves the new copy.
• Swap logos or images across an entire demo library. "Replace this old logo with this new one everywhere it appears across all my demos." Useful for company rebrands or co-branded leave-behinds.
• Personalize a demo from a sales call. "Look at the call I had with Acme yesterday and customize this demo for them." Your assistant reads the transcript (via the other connectors you've given it) and tailors the variables, guides, and content accordingly through the Reprise MCP.
• Generate guides from a script. Hand your assistant a talk track plus a captured demo, and it will write the guide text and place each guide on the right element.
• Translate a demo into another language. Bulk-translate every visible text and guide in a single prompt.
• Find the right demo to send. "Which demo in my library best fits this opportunity?" Your assistant searches across your library by content, not just by title.
• Configure Data Injection. Point your assistant at a page and it will wire up datasets for charts, tables, and dashboards — no manual response-template configuration required.
• Send product feedback directly to Reprise. Anytime the MCP can't do something you need, ask your AI assistant to "log a friction report." It goes straight into our product backlog.

New tools are being added on a near-weekly cadence. Once you're connected, ask your AI assistant "what can the Reprise MCP do?" for the latest list.

How to set it up

Connecting takes about two minutes. The Reprise MCP works with any AI client that supports the Model Context Protocol — Claude, ChatGPT, Codex, Copilot, Gemini, and others. The endpoint is the same regardless of which client you use:

Endpoint: https://app.getreprise.com/mcp/

Once you've connected, you'll be redirected to Reprise to authenticate via OAuth (inherits your existing Reprise SSO login), then sent back to your AI client. From there, try a prompt like "Use the Reprise MCP to list my recent demos" to confirm the connection.

Step-by-step instructions for the three most common clients are below.

Connecting Claude

1. In Claude (web, desktop, or Code), go to Settings → Connectors → Add custom connector.
2. Paste the endpoint: https://app.getreprise.com/mcp/
3. Click Add, then authenticate with OAuth when prompted.
4. The Reprise tools will appear in your tool list and are ready to use.

For a click-through walkthrough of the Claude setup, see this interactive demo.

Connecting ChatGPT

ChatGPT supports custom MCP servers through Developer Mode. A workspace admin needs to enable Developer Mode once; after that, individual users in the workspace can connect without it.

1. Admin (one-time): In ChatGPT, go to Settings → Connectors → Advanced settings and toggle on Developer mode. (On Business, Enterprise, and Edu plans, this can also be enabled from the admin console under Workspace Settings → Permissions & Roles → Connected Data.)
2. Go to Settings → Connectors → Create.
3. Fill in the connector details and paste the MCP server URL: https://app.getreprise.com/mcp/
4. Choose OAuth as the authentication method.
5. Click Create. ChatGPT will call tools/list and surface every Reprise tool — you can review each one before enabling.
6. Authenticate with OAuth when prompted.

Plus and Pro users can install the connector and call read-only Reprise tools. Business, Enterprise, and Edu users get full read/write capability, gated behind the admin toggle above and per-call approval cards. For more on Developer Mode and the constraints OpenAI enforces, see OpenAI's documentation.

Connecting Copilot

Microsoft Copilot Studio is the easiest place to add the Reprise MCP for Copilot users. Most enterprises will need IT to approve the connector before users can add it.

1. In Copilot Studio, open the agent you want to extend and go to the Tools page.
2. Select Add a tool → New tool → Model Context Protocol.
3. Fill in the required fields:
   • Server name: Reprise
   • Server description: Build and customize Reprise demos via natural-language prompts
   • Server URL: https://app.getreprise.com/mcp/
4. For authentication, use OAuth 2.0 with the dynamic client registration option — the Reprise MCP supports OAuth 2.0 Dynamic Client Registration (DCR), which is the simplest path.
5. Select Add and configure to finish wiring the MCP into your agent.

For more detail on the MCP onboarding wizard in Copilot Studio, see Microsoft's documentation.

A few things to know

• No new license is needed. The MCP uses your existing Reprise license. If you're a Builder, you can build via the MCP. If you're a Presenter, you can take Presenter-level actions. If you don't have a Reprise license, the MCP won't work for you.
• Your Reprise roles and permissions are respected. The MCP can only do things you specifically have permission to do in the Reprise app. Custom roles, group sharing, and folder-level permissions all carry over.
• Read-only mode is supported. If you want to give your AI assistant the ability to read your Reprise content but never write to it, request a read-only token (scope=read_only). Writes are blocked at the MCP boundary.
• HTML demos have version history. If a prompt produces an unexpected change, you can roll back to a previous version of the demo from the editor — this works the same way for MCP-driven changes as it does for manual edits. Data Injection version history is currently internal-only; if you need to roll back a dataset, contact support and we'll restore it for you. You can also export datasets to CSV as a local backup.

Questions from a security standpoint

This is the section your IT and security team will care about most.

Does the Reprise MCP add new sub-processors or new AI models? No. The MCP is hosted by Reprise on the same infrastructure as the rest of the platform, so no new sub-processors are introduced. The MCP itself contains no AI models and runs no inference — search_patterns, for example, uses BM25 lexical retrieval, not embeddings. The reasoning happens in your existing AI client (Claude, ChatGPT, etc.), which your organization has already evaluated.

Does the MCP expand what users can do in Reprise? No. The 32 tools all do things that already exist in the Reprise application and have already been scoped in our security review. We're not introducing new functionality — we're providing a new way to invoke existing functionality.

Does the MCP respect user roles and permissions? Yes. Every action runs as the authenticated user. If you don't have permission to edit a demo in the Reprise UI, you can't edit it via the MCP either. Custom permissions, builder/presenter restrictions, group sharing, and folder access all apply. Tenant isolation is enforced in three layers: an MCP scope check, request-context client binding, and an ORM client= filter on every backend query.

How does authentication work? OAuth 2.0 Authorization Code with PKCE (S256), against your existing Reprise login (SSO inherited). There's an explicit per-user consent screen. Tokens are scoped to the authorizing user and inherit that user's existing Reprise role — the MCP cannot do anything the user couldn't already do in the Reprise UI. No tokens or credentials are stored in the AI client or shared with third parties.

How are tokens revoked when a user leaves? IdP-driven deprovisioning revokes MCP tokens immediately. SCIM/IdP deactivation cascades to revoke access and refresh tokens, and the token-verify filter checks user.is_active=True on every request — there's no expiry-only window. The MCP also appears in the user's authorized-apps list and can be revoked from there in one click.

Is customer data sent anywhere new? No. The MCP communicates between your AI client and Reprise only. The data that flows through the MCP is the metadata and content for your own Reprise objects: demos, captured screens and network requests, snippets, datasets, rules, and notes. Customer internal applications, source code, production databases, internal networks, and end-user PII outside the existing Reprise tenant do not flow.

What's the egress requirement? A single host: app.getreprise.com:443 (HTTPS, TLS 1.2+).

Are there protections against prompt injection? Yes. Tool fields that surface user-captured content (html_screen, html_guide, clone_request, clone_snippet, clone_note) are wrapped in <untrusted-content>…</untrusted-content> delimiters with a data-not-instructions disclaimer, giving the agent runtime a stable contract to enforce.

Is there an audit log? Yes. State-changing MCP tool calls append to the customer-visible ClientAuditLog, stamped with mcp_request_id, mcp_client_id, and mcp_token_id. If you ever need to understand what a user did or when, the audit log captures it.

What about SOC 2, ISO 27001, and GDPR? Reprise holds SOC 2 Type 2 and ISO 27001. The MCP component is new and not yet in the most recent audit scope, but it runs in the same environment that is in scope and inherits the same operational controls. It's planned for inclusion in the next audit cycle. GDPR posture, residency, and the sub-processor list are governed by the existing Reprise customer agreement — the MCP introduces no new processor.

Need help?

If you run into trouble during setup or have questions the article doesn't cover, reach out to support at [e-mail].