Give your agents a secure data room.
Let AI agents run real document workflows on Papermark. Spin up data rooms, upload documents, mint secure links, and read page-by-page analytics over MCP and REST. 43 typed tools, scoped tokens, and a full audit trail.
Built on the developer platform · read the docs or jump to the quickstart
Your agents, operating
real data rooms.
Papermark for agents exposes the full Papermark platform to any AI agent. Instead of describing a workflow to a human, an agent calls typed tools to create rooms, share documents, and report on engagement, with the same security guarantees as the product.
- MMCP-native. Drop Papermark into Claude, Cursor, ChatGPT, or any MCP client with
@papermark/mcp-server. - AAPI-first. Every tool maps to a versioned REST endpoint, so custom agents can call it directly over HTTP.
- SSecure by default. Scoped
pm_live_tokens, link permissions, watermarking, and an append-only audit log. - GGrounded. Agents answer only from documents the token is allowed to read, with citations back to pages.
What an agent can do with Papermark.
Each card is a workflow an agent can run end to end through MCP or the REST API. Tool names map one-to-one to the 43 tools on the developer platform.
From a single instruction, an agent creates a structured data room with folders, ready for a deal or fundraise.
Agents push files into the room, set versions, and let Papermark auto-index pages for search and analytics.
Generate scoped share links with NDA gating, expiry, email verification, and dynamic watermarking on every page.
Agents pull who viewed what, time per page, and downloads to rank serious buyers and trigger follow-ups.
Search across an entire room and return answers cited back to the exact documents the agent is allowed to see.
Grant or revoke per-bidder access, rotate passwords, and keep staged disclosure tidy across a live process.
Subscribe to webhooks so an agent can respond the moment a link is viewed or a new document lands.
Pull an append-only log of every view, download, and change for the deal record or a compliance review.
Real workflows, by team.
Concrete automations teams run today through MCP, the CLI, and the REST API. Each one chains the same tools shown above into a job that used to take hours.
Founders raising a round
Automated data room setup
One prompt builds the whole room: create a Series B data room, upload a local folder, sort it into Financials, Legal, Product, and Team, and mint a password-protected, email-verified link for every VC on the list. Replaces 2-3 hours of manual UI work.
Engagement-based follow-up
After links go out, the founder asks who actually read the numbers. The agent filters views by page and dwell time and returns a ranked list, so outreach is targeted instead of guesswork.
M&A advisors and investment bankers
Per-buyer room provisioning on signed NDA
When a signed NDA lands, an agent parses the buyer's firm, creates a dedicated room or buyer group, uploads the CIM, model, and cap table from a master folder, applies a watermark with the firm name, and emails the link back. Replaces 30-60 minutes of manual SharePoint duplication per buyer.
Daily diligence digest in Slack
A scheduled job pulls yesterday's activity for each active deal and posts a morning summary to Slack. The deal team starts the day knowing exactly where every buyer stands without opening the platform.
Boutique advisors moving off Excel
Deal pipeline as a chat thread
Instead of tracking deals in a spreadsheet, the advisor asks for a deal's status and the agent pulls every active link, the latest visitor activity, and open Q&A into one summary. Adding buyers and sending the teaser is a single instruction, and every action lands in the audit trail.
Private equity and venture capital: LP communications
Quarterly LP update distribution
The GP uploads the Q4 letter and portfolio file, then mints a unique, email-gated link per LP from a CSV in one command. Two weeks later an agent flags LPs who never opened and drafts personalized nudges for review.
Slide-level analytics for IR
Page-by-page dwell times roll up across every LP view, so IR learns which parts of the deck land. "LPs spent on average 1m 40s on the track record page and 12s on the team page" feeds straight into the next deck revision.
Real estate brokers and asset sales
Listing-to-data-room automation
A broker forwards a listing email with property PDFs to a monitored inbox. The agent extracts the property, creates a room, files floor plans, financials, and inspection reports into preset folders, mints an email-gated public link, generates a QR code, and replies ready for the listing page.
SaaS platforms embedding Papermark
White-labeled rooms inside a host product
A vertical SaaS tool embeds Papermark via the REST API. When a user creates a deal or listing, the host backend provisions the room and links server-side on a custom domain. The end user never sees Papermark; the host keeps full control of the UX while Papermark handles the viewer, watermarking, NDA gating, and analytics.
Enterprise sales teams
Proposal rooms triggered from the CRM
When an opportunity hits the Proposal stage in HubSpot or Salesforce, an agent creates a link to the SOW, MSA, and pricing, sets the prospect's email as the only verified viewer, and writes the link back to the deal record. The AE never generates a proposal link by hand again.
Engagement-based AE tasks
A second workflow surfaces prospects who opened the proposal but went quiet for 5 days, ranks them by engagement depth, and creates outreach tasks for the AE automatically.
Internal analytics and compliance
Audit-log export to the warehouse
A nightly job streams every view event into Snowflake or BigQuery, so the team builds engagement, open-rate, and geography dashboards in their own BI tool, with no Papermark UI required.
Two lines to wire up an agent.
Point any MCP client at Papermark with a scoped token. Use stdio for local clients or HTTP at mcp.papermark.com for browser clients. The full reference lives on /dev.
# remote MCP server, no install required # works with Cursor, Continue, Claude Code, any MCP client { "papermark": { "url": "https://mcp.papermark.com", "transport": "http", "headers": { "Authorization": "Bearer pm_live_..." } } } # then: "create a data room called Series B and # upload every PDF in ~/acme/"
{ "mcpServers": { "papermark": { "command": "npx", "args": ["-y", "@papermark/mcp-server"], "env": { "PAPERMARK_TOKEN": "pm_live_..." } } } }
# any agent can call the REST API directly curl -X POST https://api.papermark.com/v1/datarooms \ -H "Authorization: Bearer pm_live_..." \ -H "Content-Type: application/json" \ -d '{ "name": "Series B" }'
Agents inherit your guardrails.
Automation never means losing control. Every agent action runs through the same security layer as the rest of Papermark.
Scoped tokens
Agents authenticate with pm_live_ bearer tokens you create and revoke from settings. No shared passwords, no broad access.
Permissions respected
Link-level and folder-level permissions still apply. An agent only ever reads or shares what the token is allowed to.
Watermark & protection
Dynamic watermarking and screenshot protection stay on for every link an agent creates, on every page.
Append-only audit log
Every agent view, upload, link, and change is logged immutably and exportable for the deal record.
Grounded answers
Q&A and search return responses cited back to source pages, so reviewers can verify in seconds.
Open source
Papermark is open source and self-hostable, so you can run agent workflows entirely inside your own infrastructure.
Questions about Papermark for agents
What is Papermark for agents?+
Papermark for agents lets any AI agent operate your secure data rooms over the Model Context Protocol (MCP) and a versioned REST API. Agents can create data rooms, upload and organize documents, mint secure share links, read page-by-page analytics, and react to events, all while respecting the same permissions and watermarking you use today.
What can an agent actually do with Papermark?+
An agent can run the full document workflow end to end: create a data room with folders, upload and organize documents, mint secure share links with NDA gating and watermarking, read page-by-page view analytics, run grounded Q&A across a room, manage visitor access, and export the audit log. Every one of the 43 MCP tools maps to a REST endpoint, so the same actions work from Claude, Cursor, ChatGPT, the papermark CLI, or your own code.
What are the most common agent use cases?+
The ones teams run most often: founders spinning up a fundraising data room and minting a watermarked link per VC; M&A advisors provisioning a per-buyer room the moment an NDA is signed; PE and VC firms distributing email-gated LP updates in bulk and flagging who never opened; real estate brokers turning a listing email into a ready-to-share room; SaaS platforms provisioning white-labeled rooms server-side on a custom domain; and sales teams creating proposal rooms straight from the CRM. See the "Real workflows, by team" section above for the exact tool chains.
Can an agent tell me who viewed a document and for how long?+
Yes. With get_view_analytics and list_views an agent returns page-by-page dwell time, downloads, and per-visitor activity. A typical question is "Which investors spent more than 60 seconds on the financials slide?" or "Who opened the proposal but went quiet for 5 days?", and the agent returns a ranked list you can act on or write back to your CRM.
Can agents answer questions about the documents in a room?+
Yes. search_documents runs grounded Q&A across an entire data room and returns answers cited back to the exact pages the token is allowed to see. It never invents content or reaches documents outside the agent's permissions, so reviewers can verify every answer in seconds.
Which MCP clients are supported?+
Any MCP client. Use stdio with @papermark/mcp-server for local clients like Claude Desktop and Claude Code, or connect over HTTP at mcp.papermark.com for browser clients like claude.ai Connectors and ChatGPT Apps. Same 43 tools, same shapes.
How do agents authenticate?+
With a scoped API token in the form pm_live_... created at app.papermark.com/settings/tokens. The token acts as a long-lived bearer credential and respects the same access controls as the rest of Papermark.
Do agents bypass permissions or watermarking?+
No. Agent actions run inside the same secure infrastructure as Papermark data rooms. Link-level permissions, dynamic watermarking, screenshot protection, and the append-only audit log all still apply.
Where do I find the full reference?+
The developer hub at papermark.com/dev covers the REST API, the papermark CLI, the MCP server, and webhooks. Full documentation lives at papermark.com/docs with a five-minute quickstart.
Building the standard for agents and data rooms.
Papermark is defining how AI agents securely operate data rooms over MCP and the API. Connect a client, mint a token, and build on it. Open source, hosted, or self-hosted.