Templafy MCP Server

This article explains how the Templafy MCP Server works. It connects AI assistants to your organization's on-brand document creation, powered by the same agents and templates your team already uses in Templafy. From a chat in Claude, ChatGPT, Microsoft Copilot Studio, or another Model Context Protocol (MCP) compatible assistant, end users can ask for a deck and receive a finished, on-brand PPTX  or PDF file generated by Templafy. Word documents (DOCX) are available in Public Preview.

What the MCP server does

The Templafy MCP server exposes Templafy's content-creation capabilities to AI assistants over the open Model Context Protocol standard. Once the server is connected, the AI assistant can, on the user's behalf, create presentations and — in Public Preview — Word documents.

Presentations

• Discover the right Templafy agent for the user's goal — a sales proposal, a quarterly review, a customer pitch, a status report, and so on. Agents are the curated, on-brand business document generators your organization has published in Templafy.
• List the presentation themes available in the user's Templafy environment, each with a preview image and a slide-layout count.
• Collect the deck's requirements in a short back-and-forth with Templafy's Document Agent, and prepare an outline.
• Generate a new presentation using your organization's agents, layouts, fonts, and brand rules — either through an embedded widget that runs inside the conversation, or, for text-only assistants, a step-by-step flow.
• Track the generation's progress and return download links for the finished PPTX file and a PDF copy.

The output is always a real .pptx file built from your organization's agents — never a generic AI-generated image. The content is produced by Templafy's existing content-generation platform, governed by the same controls and data boundaries you already rely on for Templafy in Office.

Word documents (Public Preview)

• The assistant writes the document section by section — for example a letter, memo, report, proposal, or scope of work. Exact content such as names, dates, amounts, tables, and addresses is kept as written; the assistant can ask Templafy to compose the remaining prose.
• Templafy applies your organization's Word template and generates the document. Word documents do not use presentation agents or themes.
• The finished DOCX file and a PDF copy are returned as download links. In assistants that support the Templafy widget, page previews are shown in the conversation; text-only assistants use a step-by-step flow.

Supported assistants

Server URL

The Templafy MCP server is hosted at a single global endpoint. There is no tenant or organization name in the URL — your Templafy organization is identified from the access token issued during sign-in.

https://mcp.templafy.com/

Use this URL anywhere an AI assistant asks for a "Server URL", "MCP endpoint", "Remote MCP server" or equivalent.

How sign-in works

The Templafy MCP server is an OAuth-protected resource. It uses standard OAuth 2.0 authorization code with PKCE, with Templafy as the authorization server — this authorizes the user with your identity in Templafy. Credentials are never shared with the AI assistant.

• The AI assistant opens a window that loads Templafy's authentication page prompting the user for an email address. After providing the email address authentication will be delegated to the tenant's configured Identity Provider. If this supports seamless authentication and the user is already signed-in with the identity provider this will complete the sign-in flow with the AI Assistant. Without support for seamless authentication additional prompts from the Identity Provider might be presented to the user.
  A user can exist in more than one tenant, in which case after providing their email address the user will be asked to select the tenant to authenticate with.

• Authentication tokens are limited to the permissions the MCP server needs. The assistant requests three scopes:

  • library.read — read your Templafy library, agents, and themes.
  • library.generate — generate presentations and documents.
  • offline_access — keep you signed in, so tokens can refresh without a new sign-in.

  This limits the impact if the AI assistant has a data leak that compromises authentication tokens.

• Tokens refresh automatically. Users only see the sign-in screen again if they revoke access or stay signed out long enough for the refresh token to expire.
• Users (or admins) can disconnect at any time from the AI assistant's connector settings. After disconnecting, the assistant should no longer call the Templafy connector for that connection.

Templafy supports two ways for AI assistants to identify themselves to the OAuth flow:

• CIMD (Client ID Metadata Documents) — the assistant publishes its own client metadata at a public URL, and Templafy fetches and validates that document at sign-in. No administrator action is required.
• Pre-registration — Pre-registration is not self-service and needs the configuration and exchange of credentials, identifiers, and scopes between your Templafy team (your account manager or solutions engineer) and the customer's AI assistant administrator or implementation partner.

Most modern AI assistants use CIMD; some enterprise platforms still require pre-registration. Both flows lead to the same Templafy sign-in window for the end user.

Add the server with CIMD

Assistants that support Client ID Metadata Documents publish their own client metadata at a public URL. Templafy fetches and validates that document at sign-in, so no admin work is needed on the Templafy side. From the user's perspective, adding the server is a two-step flow.

1. In the AI assistant, open the Connectors, Integrations, or MCP servers settings and choose Add custom connector (the exact wording varies by vendor).
2. Enter the Server URL https://mcp.templafy.com/, then click Connect (or equivalent). A Templafy sign-in window opens; sign in with your Templafy account and approve the requested scopes.

That's it. The assistant is now connected and the Templafy tools are available in your conversations.

Add the server with pre-registration

Some enterprise AI assistants — typically those built on identity providers that require known OAuth clients — need a pre-registered OAuth client rather than CIMD. Pre-registration is not self-service: the configuration and exchange of credentials, identifiers, and scopes is handled directly between your Templafy team (your account manager or solutions engineer) and the customer's AI assistant administrator or implementation partner. Once that exchange is done, users sign in normally.

The exchange typically runs as follows.

1. The customer's AI assistant administrator (or their implementation partner) contacts their Templafy account manager to request pre-registration, and provides:
   • A display name for the assistant (e.g. Internal Copilot).
   • The redirect URL the assistant will use. The assistant's documentation publishes this — it is the URL its OAuth flow will return to after sign-in.
   • The expected client type (public or confidential), and the contact who will receive the credentials.
2. The Templafy team registers the assistant on the Templafy side and returns, through a secure channel agreed with the customer:
   • A Client ID.
   • A Client secret, only for confidential-client setups.
   • Confirmation of the registered redirect URL and granted scopes (library.read, library.generate, and offline_access).
3. The customer's AI assistant administrator configures the assistant with:
   • Server URL: https://mcp.templafy.com/
   • Client ID: the value returned by Templafy
   • Client secret: only if the assistant is a confidential client and Templafy issued one
   • Scopes, if the assistant asks for them: library.read library.generate offline_access. Without offline_access, users have to sign in again each time their access token expires.
4. End users connect from the assistant's connector settings as in the CIMD flow. The first time they sign in, they approve the same Templafy scopes.

Which setup path do I use?

Try the CIMD flow first — if the assistant accepts https://mcp.templafy.com/ and walks the user through Templafy sign-in without asking for a Client ID or Client secret, you're on CIMD and you're done. If this is not supported switch to the pre-registration flow and get in contact with your Templafy team (your account manager or solutions engineer) and the customer's AI assistant administrator or implementation partner.

Claude setup

Claude supports custom connectors using remote MCP. On Team and Enterprise plans, an Owner or Primary Owner adds the connector to the organization first; members then connect their own Templafy account. On individual plans, users can add the custom connector from their own connector settings.

1. In Claude, open Customize → Connectors. On Team and Enterprise plans, an Owner can instead open Organization settings → Connectors.
2. Click Add, choose Browse connectors then search for Templafy.
3. Click the Templafy entry.
4. Finish adding the connector by clicking Connect to Claude, sign in with Templafy.

Claude connects to remote MCP servers from Anthropic's cloud infrastructure, so https://mcp.templafy.com/ must be reachable from the public internet. Inside Claude.ai, the Templafy MCP server can provide embedded widgets that render directly in the conversation. Claude Code uses a text-mode equivalent flow instead of the widgets.

ChatGPT setup

ChatGPT can connect to remote MCP servers as Plugins. For organization-wide rollout, a workspace admin should control whether plugins are allowed and how they are made available.

1. In ChatGPT, open Plugins.
2. Search for Templafy under search plugins.
3. Click the + sign on the Templafy entry.
4. Finish adding the plugin by clicking Sign in with Templafy.
5. Open a new chat, add the Templafy app from the message composer tools, and sign in with Templafy when prompted.

When the plugin is available in a chat, ask for a presentation or, if enabled for your tenant, a Word document in natural language. ChatGPT may ask you to approve tool calls before data is sent to the Templafy MCP server, especially for tools that create content.

Microsoft 365 Copilot Chat and Copilot Cowork

For Microsoft 365 Copilot Chat and Microsoft Copilot Cowork, you don't add the MCP server yourself. Install the Templafy app for Microsoft 365 instead: it connects both surfaces to the Templafy MCP server and handles sign-in. See How to use the Templafy App for Microsoft 365 in Teams, Copilot Chat & Copilot Cowork.

• Copilot Chat creates presentations. Word document creation is not supported in Copilot Chat during Public Preview.
• Copilot Cowork creates presentations and, once enabled for your tenant, Word documents (Public Preview).

Use Microsoft Copilot Studio (below) only if you are building your own agent and want to add the Templafy MCP server to it.

Microsoft Copilot Studio setup

Microsoft Copilot Studio can add an existing MCP server to an agent. Authentication is configured through Power Platform connector infrastructure, and the available OAuth setup path depends on what the assistant and identity provider support. If Copilot Studio asks for a Client ID, Client secret, authorization URL, token URL, or callback URL, follow the pre-registration path with your Templafy team.

1. In Copilot Studio, open the agent and go to Tools.
2. Select Add a tool, then choose the option to add an existing MCP server.
3. Enter the MCP server URL https://mcp.templafy.com/.
4. Select the authentication type. For OAuth 2.0, use the most automated option available first; if Copilot Studio requires manual OAuth details, request pre-registration from your Templafy team and copy the callback URL into that registration.
5. Create a new connection or use an existing one, then add the MCP server to the agent.
6. Publish or test the agent. End users sign in with Templafy the first time the connector is used.

Generic MCP consumer setup

Any MCP-compatible assistant that supports remote MCP servers can use the Templafy MCP server. The setup is always the same shape: add a connector, enter https://mcp.templafy.com/, and complete OAuth sign-in.

1. In the assistant, find the area named Connectors, Integrations, Tools, Apps, or MCP servers.
2. Choose Add custom connector, Add MCP server, or the equivalent option.
3. Use these values where the assistant asks for them:
   • Name: Templafy
   • Server URL, endpoint, or connector URL: https://mcp.templafy.com/
   • Authentication: OAuth 2.0 / OAuth, if the assistant asks
4. Complete Templafy sign-in and approve the requested scopes.
5. If the assistant asks for Client ID, Client secret, callback URL, redirect URI, authorization URL, token URL, or scopes before it can connect, follow the pre-registration path.

If your assistant of choice is not listed above, contact Templafy support with the assistant name and we will help confirm the right setup path.

Available tools

Once connected, the assistant can call the following Templafy tools on the user's behalf. The user does not need to call them explicitly — the assistant picks the right one for the conversation.

Presentations

Word documents (Public Preview)

These tools are only available when Word document creation is enabled for your tenant.

Example prompts

Once the Templafy MCP server is connected, your users can simply describe the document they need. Some prompts to copy-paste and adapt:

Presentations

• Prepare a 10-slide Q3 business review for our EMEA leadership meeting. Cover revenue, pipeline, top wins, and risks. Use our standard corporate theme.
• Build a customer-facing pitch deck introducing our new platform release. Audience is a CIO and CTO at an enterprise prospect. Use the Sales agent if we have one.
• Generate a board update on our FY26 strategy: opening summary, three strategic priorities with a slide each, financial outlook, and a closing ask.
• Which presentation themes are currently available in my Templafy environment? Show their names and previews.

Word documents (Public Preview)

• Write a one-page letter to Acme Corp confirming the renewal of their contract from 1 January at the agreed annual fee of EUR 120,000.
• Draft a scope of work for a six-week onboarding project: objectives, deliverables, timeline, responsibilities, and assumptions.
• Create an internal memo announcing our new travel policy, effective 1 November, with a short summary and the key changes as a table.

When generation completes, the assistant shows a download link in the conversation — open it to get the finished file, already on your organization's brand.

Data handling & privacy

• The Templafy MCP server only invokes Templafy's own first-party services. It does not call third-party AI providers directly; any LLM calls happen inside Templafy's existing data boundary, governed by your existing Data Processing Agreement.
• Prompts and document content sent are processed by Templafy's document generation services in order to produce the requested file. The MCP server itself stores no user prompts or content — all session state is held in Templafy's existing content-generation platform.
• The MCP server only reads the data it needs to fulfill the request: the user's prompt and document content, the organization's published agents and themes, and the status of in-progress generations the user has initiated.
• For full details on data collected, use, storage, retention, sub-processors and your rights as a data subject, see the Templafy privacy policy: Templafy privacy policy.
• Disconnecting the connector in the AI assistant immediately revokes its access.

Known limitations

• Best experience inside the AI assistant's main UI. Assistants that render rich connector UI give the smoothest experience because the Templafy widgets can run inline. Assistants that only use plain text work for short prompts but may time out on long generations. If you hit a timeout, simplify the prompt or try the same prompt in the assistant's web UI.
• One organization per connection. A connection is bound to the Templafy organization of the signed-in user. To use the connector against a different organization, disconnect and reconnect with the matching account.
• Word documents are in Public Preview. Word document creation must be enabled for your tenant, and behavior may change before general availability.
• No Word documents in Microsoft 365 Copilot Chat. Word document creation is not supported in Copilot Chat during Public Preview, even when it is enabled for your tenant. Use Copilot Cowork or another supported assistant.

For help with common issues, see Troubleshooting Templafy MCP Server.