Skip to content
MCP-native · Agent tool calls
CursorPortEden

Secure Your Data With Cursor

Cursor's Agent can call MCP tools, so it reaches well beyond your repo into email, Drive, calendars, and tickets. Add PortEden as a remote MCP server and Cursor reads those sources with PII redaction, least-privilege scope, and a full audit log, with no raw tokens in your editor.

See pricing

Add to ~/.cursor/mcp.json · 5-minute setup · Free to start

Works With What You Already Use
Gmail
Gmail
Outlook
Outlook
Google Calendar
Google Calendar
Google Drive
Google Drive
Google Docs
Google Docs
OneDrive
OneDrive
SharePoint
SharePoint
Teams
Teams
Slack
Slack
Notion
Notion
Asana
Asana
Monday
Monday
Linear
Linear
Jira
Jira
Confluence
Confluence
Entra ID
Entra ID
Better Together

What Cursor's Agent Reaches by Default

  • Cursor's Agent calls MCP tools on your behalf. Cursor's own docs warn that MCP servers can access external services and execute code on your behalf, and a typical email or drive server hands the Agent broad, un-redacted access.
  • Wiring a raw Gmail or Microsoft OAuth token (or a provider API key) into a local mcp.json or env var puts a long-lived credential on your machine, and for project configs, potentially in your repo.
  • Cursor's Privacy Mode and zero-retention agreements cover the code and prompts sent to the model provider. They do not govern what a third-party tool does with the data it returns. That layer is on you.
  • There is no built-in, exportable record of which message, file, or ticket the Agent opened through a tool, or what it received back.
Capability Matrix

Cursor With PortEden: What You Get

PII redaction before the Agent reads tool output
50+ identifier types stripped or tokenized in the response
CursorNo
+ PortEdenYes
Per-folder, per-contact, per-file scope
CursorNo
+ PortEdenYes
No raw OAuth token in mcp.json or env
PortEden holds the provider token; Cursor sends only a PortEden key or OAuth session
CursorNo
+ PortEdenYes
Exportable, per-call audit log
CursorNo
+ PortEdenYes
Read-only enforcement on send, delete, share
Cursor asks approval, but the underlying tool still can write
CursorPartial
+ PortEdenYes
Focused, token-scoped tool set
Cursor limits how many tools the Agent sees at once; PortEden exposes a small set per token
CursorN/A
+ PortEdenYes
Revoke access without touching your sign-in
CursorNo
+ PortEdenYes
Works in both global and project MCP config
CursorN/A
+ PortEdenYes
MCP Server Coverage

A Remote MCP Server, Same Firewall as Claude and ChatGPT

Connection: Remote MCP server in ~/.cursor/mcp.json (or project .cursor/mcp.json)

Email (Gmail + Outlook)

The Agent searches and reads mail through PortEden, redacted and audited. Send and reply stay scoped to what the token allows.

Google Drive + Docs

File search and read with file-level rules. Downloads return links, never raw bytes.

Google Calendar

Events and free/busy with attendee redaction, so scheduling context never leaks full attendee data.

Slack, Notion, tickets

Pull a thread, a doc, or an issue into the Agent's context through PortEden's MCP servers, with the same controls.

Tooling Notes
  • Cursor supports remote MCP servers via a url plus headers in ~/.cursor/mcp.json (global) or .cursor/mcp.json (project), and supports OAuth for remote servers.
  • Add PortEden's remote endpoints (mcp.porteden.com/email, /calendar, /drive). The Agent discovers the tools and, by Cursor's default, asks for approval before using them.
  • Cursor caps how many tools the Agent sees at once, so a focused, token-scoped tool set matters. PortEden exposes a small set per token rather than dozens.
  • Cursor's Privacy Mode governs model-provider traffic, not tool data. PortEden is the control layer for what flows through the tools.
  • The provider OAuth token never enters your editor. Cursor holds only a PortEden key or OAuth session.

Add PortEden to Cursor in Three Steps

1

Connect a source in PortEden

Sign in to PortEden and connect Gmail, Outlook, Drive, or Calendar. PortEden holds the OAuth token and gives you a scoped MCP endpoint.

2

Add it to mcp.json

Add a remote server entry to ~/.cursor/mcp.json (global) or .cursor/mcp.json (project) with PortEden's url and an auth header or OAuth. Enable it in Settings, Features, Model Context Protocol.

3

Scope it and use it

Set the token's scope and redaction in PortEden, then ask the Agent to work across your data. Approve the tool calls; every call is redacted and logged.

Cursor + PortEden

Five-Minute Setup. Free While You Test.

Connect a data source, plug Cursor into PortEden, and put Cursor to work on the data your team actually needs to handle.

MCP setup guide

Frequently Asked Questions

Does Cursor support MCP servers?
Yes. Cursor reads MCP servers from ~/.cursor/mcp.json (global) and .cursor/mcp.json (project), supports local stdio and remote SSE or Streamable HTTP servers, and supports OAuth for remote servers. PortEden is added as a remote server.
Does Cursor or its model see my raw OAuth token?
No. The Google or Microsoft token stays inside PortEden. Cursor's config holds only a PortEden key or OAuth session, and the Agent gets back redacted, scoped results.
Cursor already has Privacy Mode, so why add PortEden?
Privacy Mode and zero-retention agreements cover the code and prompts Cursor sends to the model provider. They do not govern what a third-party MCP tool does with the data it returns. PortEden is the control layer for that tool data: redaction, scope, and audit on every call.
Can the Agent send email or delete files?
Cursor asks for approval before using MCP tools by default, but the underlying tool still has whatever the token allows. Keep the PortEden token read-only until a workflow is proven, then widen scope deliberately.
What does PortEden see of my Cursor session?
Only the tool calls the Agent makes through the firewall: the request, the access-rule decision, and the redacted result. PortEden does not see your code, your prompts, or the model's output.
Should I use the global or the project config?
Both work, and if the same server name appears in both, the project config takes priority. Keep secrets out of committed project configs: prefer the global ~/.cursor/mcp.json or an environment-injected header so a token never lands in your repo.
How do I revoke access?
Remove the server from mcp.json, or revoke the token and disconnect the source in PortEden. Either cuts the Agent off without affecting your own sign-in.
What does it cost?
PortEden is free to start. Higher tool-call quotas, SSO, and SIEM export are on paid plans. See pricing for details.

Get More From Cursor With PortEden

Five-minute setup. Free tier for solo licensed practitioners. Same AI you already use — now ready for the work your team actually needs to do.

Talk to sales

Rolling out to a whole team? Talk to sales →