• CLEAN
  • SAFE
  • LOGGED
  • HUMAN

How to redact sensitive data from your AI tool requests

Connect MintMCP once to add a privacy layer between your agent and its external tools. It intercepts outgoing requests to strip out sensitive information automatically.

Ask AI about this page

Short answer

How can I stop my AI from sending PII to other tools?

MintMCP acts as a gatekeeper that scans your agent's requests for sensitive strings like emails or credit card numbers. It scrubs that data before the request ever hits a downstream tool, ensuring your private info stays local. You get a clean, safe request every time.

Privacy outcomes.

Where your data lands.

The MCP handles the heavy lifting of data sanitization.

CLEAN

Sanitized requests

The tool replaces sensitive strings with generic placeholders. Your downstream tools receive valid but anonymous data.

SAFE

PII protection

Automated checks catch emails, phone numbers, and keys. This prevents accidental data leaks to third-party APIs.

LOGGED

Audit trails

You can see what was caught and what was passed through. This gives you visibility into your privacy posture.

HUMAN

Manual review

If a guardrail triggers a high-risk alert, the process stops. You decide whether to proceed or fix the prompt.

The workflow

What your AI does when a request is made.

The MCP intercepts the communication flow to inspect the payload.

  1. Intercept request

    The MCP catches the outgoing tool call from your agent.

    mintmcp_eval_guardrail

  2. Scan for PII

    The guardrail checks the text against patterns for sensitive data.

    mintmcp_eval_guardrail

  3. Redact strings

    Sensitive values are swapped for safe tokens to protect privacy.

    mintmcp_eval_guardrail

  4. Execute action

    The sanitized request is sent to the final destination.

    mintmcp_run_tool_action

Try it

Copy these to start.

Use these prompts to test your privacy setup.

Starting points

These are basic ways to trigger the guardrails. Swap out the specific data types to match your needs.

MintMCP Connector

You're all set. Choose your MCP client and follow the setup instructions.

Connector linkhttps://edge.vinkius.com/vk_preview_O150J9v0kxg8rdyosRWdbRwIt3xDeHzZmIR4xy76/mcp

Claude Desktop

Follow the steps below to connect in seconds.

  1. 1In Claude Desktop, open Settings → Connectors.
  2. 2Click “Add custom connector” and paste the connector link above as the remote MCP server URL.
  3. 3Click Add and start a new chat — MintMCP capabilities are ready to use.
Configuration · claude_desktop_config.jsonCopy
{
  "mcpServers": {
    "mintmcp-mcp": {
      "url": "https://edge.vinkius.com/vk_preview_O150J9v0kxg8rdyosRWdbRwIt3xDeHzZmIR4xy76/mcp"
    }
  }
}
Copy into chat04

  • Check this customer list for any PII before sending it to the CRM.

  • Redact all email addresses from this support ticket before processing.

  • Scrub any credit card numbers found in this transaction log.

  • Clean this Slack message of sensitive user data before it hits the database.

  • Claude
  • ChatGPT
  • Cursor
  • VS Code
  • Windsurf
  • Claude Code
  • JetBrains
  • Cline

Start here

Connect MintMCP once, then ask.

Just link your account once. Your credentials stay encrypted, and you can immediately start using the guardrails in your preferred client.

Connect MintMCP to your AI

FAQ

How the redaction behaves

  • 01

    Does the AI delete my original data?

    No. The redaction happens in transit. Your original data remains in your local context, but the version sent to the tool is scrubbed.

  • 02

    Can I choose which types of data to redact?

    Yes. You can instruct the agent to look for specific patterns or use the built-in guardrails for common PII.

  • 03

    Will the tool break my downstream applications?

    It depends on the tool's requirements. If a tool requires a real email address to function, redacting it will cause that tool to fail.

  • 04

    Does MintMCP store my sensitive information?

    No. The MCP processes the data to redact it and does not persist the sensitive strings it finds.

  • 05

    Can the AI bypass the guardrails?

    No. The guardrails are designed to intercept the tool call itself, so the agent cannot bypass them by simply asking to ignore them.

  • More questions about MintMCP? The Connector page answers them. See everything the MintMCP Connector can do

Connect MintMCP to Claude, Cursor, ChatGPT & more