Access & permissions
The Scrive MCP has a simple permission model: it acts as you. When you connect it, you sign in to Scrive and approve access, and every tool call runs with a token issued for your user. The agent can see and do what you can in the user interface, minus anything the tools don’t expose.
Authentication
Section titled “Authentication”The Scrive MCP at https://mcp.scrive.com/mcp is an OAuth 2.0 protected resource. Connecting it works the same way in every AI client:
- You add the server URL to your client.
- The client discovers Scrive’s authorization server,
https://oauth2.scrive.com, from the server’s metadata and opens a browser window. - You sign in to Scrive, if you are not already, and approve the request.
- The client receives an access token for your user, which it sends with every tool call.
Most AI tools need nothing more than the server URL, and there are no API keys to copy. A few tools also ask for a client ID and secret; see When you need a client ID and secret. Tokens are held by your AI client, not by Scrive’s MCP server, which forwards each request’s token to the Scrive API and keeps nothing between requests. The single scope, full, grants the same access your user has in the Scrive app.
When you need a client ID and secret
Section titled “When you need a client ID and secret”A client ID and secret identify the AI tool to Scrive, not you. Everyone still signs in with their own Scrive account, and the agent acts only as them. Most AI tools identify themselves to Scrive automatically when you add the server URL. The tools below cannot, and need a client ID and secret issued by Scrive. An admin sets them up once for the organisation.
| AI tool | Where you enter them | Callback URL to send to Scrive |
|---|---|---|
| Copilot Studio | MCP wizard, OAuth 2.0 with Type: Manual | Shown in the MCP wizard after you select Create |
| Slackbot | Slack app, MCP Servers, Auth type: Manual OAuth | https://oauth2.slack.com/external/auth/callback |
| Gemini Enterprise | Add MCP Server, OAuth 2.0 | https://vertexaisearch.cloud.google.com/oauth-redirect |
Request credentials from Scrive support
Section titled “Request credentials from Scrive support”Contact Scrive support and include:
- the AI tool you are connecting, for example Slackbot or Gemini Enterprise
- its callback URL from the table above (some tools call it the redirect URL)
- a contact email, the address that receives the credentials
Receive the credentials
Section titled “Receive the credentials”Scrive support sends a link to the contact email. The link shows the client ID and secret and can be opened only once, so open it when you are ready to set up the tool and copy both values straight into the tool’s settings. If the link was already opened or you lost the secret, contact Scrive support for new credentials.
Enter them in your AI tool
Section titled “Enter them in your AI tool”Tools that ask for more than the client ID and secret also need:
| Field | Value |
|---|---|
| Authorization URL | https://oauth2.scrive.com/oauth2/auth |
| Token URL | https://oauth2.scrive.com/oauth2/token |
| Scope | full |
| PKCE | On |
Treat the secret like a password: enter it only in the AI tool’s settings and never paste it into a chat or email. If it leaks, ask Scrive support to replace it.
User permission model
Section titled “User permission model”The MCP server does not grant any new permissions. An agent acting with your token can only do what you can already do in Scrive:
- It sees the documents your user can see in the Scrive app, and nothing else.
- Documents and bundles it creates are owned by you and appear in your account.
- The eID methods and delivery channels it can set are those enabled for your account. Setting a method that is not enabled fails when signing starts.
- It cannot change account, company or user settings.
What the agent can and cannot do
Section titled “What the agent can and cannot do”| Can | Cannot |
|---|---|
| List, filter and search documents and bundle / flow drafts | Move documents to the trash or delete them |
| Read a document’s full details and download its PDF | Change account, company or user settings |
| Upload PDFs and create documents from templates | Sign a document on someone else’s behalf |
| Edit documents in preparation, add and update parties | Act outside your own Scrive permissions |
| Start signing, send reminders, cancel pending documents | |
| Create, fill, start and delete bundle / flow drafts | |
| Read usage statistics |
Rate limits
Section titled “Rate limits”The Scrive MCP allows roughly ten tool calls per twenty seconds per user. Beyond that it returns an error asking the agent to wait, and the agent retries after the pause. Bulk work, such as reminding a hundred documents, takes a few minutes for this reason.
Data processing
Section titled “Data processing”Data read through the MCP (document details, party names, emails and personal numbers, excerpts of document text) is passed to your AI client and its model provider. Review your AI provider’s privacy policy and your organisation’s rules before connecting. Scrive receives the API calls made on your behalf, the same as when you use the app.
Audit trail
Section titled “Audit trail”Everything the agent does is done as your user. Documents it creates, parties it adds and processes it starts show in Scrive’s document history exactly as if you had done them in the app.
Revoking access
Section titled “Revoking access”Disconnect the Scrive connector in your AI client; that discards its tokens. See Revoke access.