# Data Sanitizer for Privacy MCP for AI Agents AI Agent Connect > Data Sanitizer for Privacy MCP scrubs sensitive PII from your text before it ever reaches an LLM. It uses deterministic redaction to swap out emails, phone numbers, and SSNs for safe placeholders, ensuring your data stays private while keeping the context intact for your AI client. ## Overview - **Category:** security - **Price:** Free - **Endpoint:** https://edge.vinkius.com/vk_preview_rcPaGicObX9YPls6pzAQexfhnCO5iETbV7M06gVi/ai-agent-connect - **Tags:** pii, redaction, compliance, security, privacy ## Description Sending sensitive customer data or personal details to an AI client is a massive risk for any business. You want the intelligence of a large language model, but you can't risk leaking social security numbers, private emails, or phone numbers into a training set or a third-party log. This MCP solves that problem by acting as a privacy shield between your raw data and your agent. Instead of manually hunting through documents to delete names or numbers, you let the tool handle the heavy lifting. It identifies sensitive entities and replaces them with consistent placeholders so the logic of your text remains perfectly intact. You can choose to just scan a block of text to see how much sensitive info is hiding in there, or you can go a step further and completely scrub the content. By using this through the Vinkius catalog, you get a reliable way to maintain compliance without slowing down your development or research workflows. It's about making sure your AI interactions are smart without being reckless with your most sensitive information. ## Tools ### get_redaction_summary Provides an overview of the supported redaction patterns ### sanitize_text Swaps sensitive info for placeholders. This is the main way to clean your data for safe use. ### validate_entity_type g., "EMAIL") to check its validity. Checks if a specific entity type is supported by the sanitizer ## Prompt Examples **Prompt:** ``` Can you sanitize this text: 'My email is john.doe@example.com and my phone is 555-0199'? ``` **Response:** ``` My email is **[EMAIL_REDACTED]** and my phone is **[PHONE_REDACTED]**. ``` **Prompt:** ``` Check if there is any sensitive data in: 'The user's SSN is 000-00-0000.' ``` **Response:** ``` The text contains one sensitive entity of type **SSN**. ``` **Prompt:** ``` Sanitize this sentence using full sentence redaction: 'Contact me at 555-1234 immediately.' ``` **Response:** ``` **[REDACTED]** ``` ## Capabilities ### Scrub PII from text Replace sensitive details like emails and SSNs with safe, consistent placeholders. ### Scan for sensitive data Check a block of text to see how much private information is present without changing it. ### Redact entire sentences Wipe out full lines of text to prevent any context leakage of private information. ### Identify supported entities See exactly which types of sensitive data the tool is trained to detect. ### Maintain context through placeholders Use deterministic redaction so the AI understands the structure of the data without seeing the actual values. ## Use Cases ### Cleaning customer support logs A support lead needs to summarize chat logs but can't send customer phone numbers to an AI. They use sanitize_text to scrub the logs first. ### Auditing medical research notes A researcher wants to use an agent to find patterns in notes but must hide patient SSNs. They use audit_text_sensitivity to check the files. ### Automating legal document review A paralegal uses the MCP to redact sensitive names and addresses in contracts before asking an agent to summarize the terms. ### Securing developer workflows A dev wants to paste error logs into an agent but realizes they contain user emails. They sanitize the logs to stay safe. ## Benefits - Stop leaking PII by using sanitize_text to swap real emails and phone numbers for safe placeholders. - Reduce compliance headaches by using audit_text_sensitivity to check for data risks before you send them. - Keep your AI's logic intact by using deterministic redaction instead of just deleting everything. - Prevent context leakage by using full sentence redaction when entity-level scrubbing isn't enough. - Know exactly what you can protect by using get_supported_entities to see the list of detectable types. ## How It Works The bottom line is you get a way to use sensitive data with AI without actually exposing it. 1. Connect the MCP to your preferred AI client via Vinkius. 2. Provide the text you want to process to your agent. 3. Receive cleaned, privacy-compliant text ready for processing. ## Frequently Asked Questions **How can I protect my data when using the Data Sanitizer for Privacy MCP?** You use the tool to replace sensitive information like emails or SSNs with placeholders before the text is sent to your AI client. **Will the Data Sanitizer for Privacy MCP change the meaning of my text?** No, it uses deterministic redaction to swap sensitive values for labels, so the context and structure remain for your agent to understand. **Can I use the Data Sanitizer for Privacy MCP to just check for risks?** Yes, you can scan text to see if it contains sensitive information without actually modifying the content. **Does the Data Sanitizer for Privacy MCP work with any AI client?** Yes, it works with any MCP-compatible client like Claude, Cursor, or Windsurf. **What kind of sensitive information can the Data Sanitizer for Privacy MCP find?** It can detect common PII including emails, phone numbers, and social security numbers. **What kind of sensitive data can be sanitized?** The tool can detect and redact various entity types including EMAIL, PHONE, SSN, and CREDIT_CARD. You can check all available types using the `get_supported_entities` tool. **What is the difference between entity and sentence redaction?** Entity redaction replaces only the specific sensitive string. Sentence redaction replaces the entire sentence containing the sensitive data to prevent leakage through surrounding context. **How can I check if a text contains sensitive information without changing it?** You can use the `audit_text_sensitivity` tool to get a report on the presence and density of sensitive entities without modifying the original text. **How does the redaction work?** The tool uses deterministic regex patterns to find sensitive data and replaces it with a placeholder like `[EMAIL_REDACTED]`. You can also choose to redact the entire sentence containing the sensitive data. **What entity types can be redacted?** You can redact various types including EMAIL, PHONE, SSN, and CREDIT_CARD. Use `get_redaction_summary` to see the full list of supported types. **Can I redact whole sentences?** Yes, by setting the `redactEntireSentences` parameter to true in the `sanitize_text` tool, the entire sentence containing the sensitive information will be replaced.