MCP security
Last updated: 13 September 2026
DocStash is a remote MCP server (Streamable HTTP over HTTPS) that lets your AI assistant create, save, and share documents in your DocStash account. This page describes exactly how it authenticates, what its tools can do, and how your data is handled.
Endpoint:https://mcp.docstash.ai
Authentication
Every connection is secured with the MCP OAuth 2.1 standard, including server metadata discovery (RFC 8414) and Dynamic Client Registration (RFC 7591). Your MCP client runs the sign-in in your browser, so the server never sees your Google or SSO password. Tokens are user-scoped and revocable at any time from your DocStash settings or your client. A token only ever acts as you, on your own account.
What the tools can do
The server exposes a fixed, published set of tools. The live schema is discoverable by any client via the standard MCP tools/list handshake at the endpoint above.
The full contract (every tool description, input schema, the always-on server instructions, and what each tool returns) is mirrored publicly at github.com/docstash/mcp-contract.
Create
Author a new document of the given type.
create_pagecreate_appcreate_pdfcreate_docxcreate_sheetcreate_textEdit
Modify the contents of an existing document of the given type.
edit_pageedit_appedit_pdfedit_docxedit_sheetedit_textRender
Render an existing document and show it to you as a live preview. The content is not read into the assistant.
get_pageget_pdfget_docxget_sheetget_textRead
Read a document's content into the assistant's context, and list your documents.
read_documentlist_documentsScreenshot
Render a screenshot image of one of your documents.
screenshot_documentSave & manage
Save a preview as a document, trash or restore it, and control who it is shared with.
stashdiscardmanage_documentmanage_sharingAccount
Read your organizations and members, and switch the active organization.
get_organizationsset_organizationlist_org_membersData handling
Private by default
Documents are private to you unless you explicitly make one public or grant someone access.
When an AI agent is connected
When you authorize an agent (Claude, ChatGPT, Cursor, etc.) we store the agent's client name and the device that authorized it, so you can revoke it. The agent then acts only as you, on your own account: the documents it creates, edits, or reads are handled exactly like documents you make yourself, and its requests appear in our server logs like any other. We never receive the agent's conversation with you, only the specific tool inputs it sends us, such as the content of a document you asked it to create.
Full details
The complete data practices live in our Privacy & Terms.