How to use the Papermark MCP server with Claude, Cursor, and ChatGPT?

The Papermark MCP server exposes the full Papermark API to any MCP-compatible AI client. With it, an assistant like Claude or Cursor can create data rooms, upload documents, generate links, and read analytics on your behalf, using 43 typed tools.

The MCP server is available on the Business plan and above. Self-hosted Papermark instances can connect AI clients on any plan.

What is MCP?

MCP stands for Model Context Protocol. It is an open standard for connecting AI assistants to external tools and data sources. An MCP server gives an AI client a set of typed tools it can call, instead of you copy-pasting context.

Papermark ships an official MCP server, @papermark/mcp-server on npm, plus a hosted HTTP endpoint at mcp.papermark.com/mcp. Both expose the same 43 tools.

What the AI can do

Once connected, an AI assistant can:

• Create and manage data rooms: create_dataroom, list_datarooms, update_dataroom
• Upload documents: upload_document from a file path or URL
• Mint share links: create_link with password, expiry, email gate, download permission
• Search across your content: search_documents by query
• Pull analytics: list_link_views, get_view_analytics for page-by-page durations
• Manage folders, visitors, and permissions through the rest of the tools

A typical agent prompt: "Create a data room called Series B, upload every PDF in ~/acme/, then mint a password-gated link that expires in 30 days."

Step 1: Generate an API token

1. Log into Papermark.
2. Go to Settings, then API Tokens.
3. Click Create token and pick the scopes you want the AI to have.
4. Copy the token. It starts with pm_live_.

For AI assistants, consider creating a token with limited scopes (for example, no delete permissions) so the agent can't accidentally remove data.

Step 2: Connect Claude Desktop or Claude Code (stdio)

For local AI clients like Claude Desktop and Claude Code, use the stdio transport.

1. Open your Claude Desktop config file:
   • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
   • Windows: %APPDATA%\Claude\claude_desktop_config.json
2. Add Papermark to mcpServers:

{
  "mcpServers": {
    "papermark": {
      "command": "npx",
      "args": ["-y", "@papermark/mcp-server"],
      "env": {
        "PAPERMARK_TOKEN": "pm_live_..."
      }
    }
  }
}

3. Save the file and restart Claude Desktop.
4. In a new chat, ask Claude something like "list my Papermark documents". You should see Claude call the MCP tool and return results.

The same mcpServers config works in Claude Code.

Step 3: Connect Cursor, claude.ai Connectors, ChatGPT Apps (HTTP)

For browser-based and other HTTP-capable clients, use the hosted MCP endpoint.

{
  "papermark": {
    "url": "https://mcp.papermark.com/mcp",
    "transport": "http",
    "headers": {
      "Authorization": "Bearer pm_live_..."
    }
  }
}

No install required. Drop this into your client's MCP settings and you're connected.

Supported HTTP clients

• Cursor (Settings, MCP)
• claude.ai Connectors (web)
• ChatGPT Apps
• Continue
• Any other MCP-compatible client

OAuth 2.1 device flow

Instead of pasting a long-lived token, you can sign in interactively. The MCP server walks you through a browser-based device-code login and stores the token at ~/.config/papermark/auth.json. No copy-paste credentials in chat logs.

npx @papermark/mcp-server login

The 43 tools at a glance

The server exposes one tool per API operation. A few highlights:

See the full list in the @papermark/mcp-server README or in tools/list once the server is connected.

Example agent prompts

• "Create a data room called Acme Series B with folders for Financials, Legal, and Diligence."
• "Upload ~/Downloads/pitch.pdf to the Acme Series B data room and mint a password-gated link that expires in 14 days."
• "Which investor spent the most time on page 7 of our pitch deck this week?"
• "Summarize the engagement on link lk_8K2m and identify visitors who only viewed the first three pages."
• "Pull the page-by-page durations for view vw_3m9k and tell me where the visitor dropped off."

Security and audit

• The MCP server uses your API token, so it inherits the token's scopes.
• All tool calls go through the standard API and are logged to your team's audit log.
• You can revoke the token at any time from Settings, API Tokens.
• Tool calls are not stored by Papermark beyond standard request logging.

Frequently asked questions

Related guides

• How to use the Papermark REST API
• How to use the Papermark CLI
• AI agent for a single document
• AI agent for an entire data room

Need help getting connected? Contact support@papermark.com or use the in-app chat.