Strac MCP Dlp已验证已精选

What is Strac MCP DLP?

Strac MCP DLP is an open-source MCP server that lets any AI agent — Claude, Cursor, VS Code, or your own — find and strip sensitive data out of text and files. It calls the Strac DLP API, so classification runs server-side and no detection model ships in the package.

It covers 191 built-in data elements across 10 categories: personal data, health data, payment data (card numbers, CVV, IBAN, routing numbers), national IDs, passports and driver licences across ~60 countries, and credentials such as AWS keys, GitHub and GitLab tokens, GCP and Azure secrets, private keys and database connection strings.

How to use Strac MCP DLP?

Install the package:

pip install strac-mcp-dlp

Then add it to your MCP client with your API key:

{
  "mcpServers": {
    "strac-dlp": {
      "command": "strac-mcp-dlp",
      "env": { "STRAC_API_KEY": "sk_live_your_key_here" }
    }
  }
}

Restart your client and ask it to redact something: "Use Strac to redact this before I paste it into the ticket."

Request a key at https://www.strac.io/mcp-integrations. Keys are prefixed sk_live_ (production) or sk_test_ (sandbox), and the server selects the matching endpoint automatically.

Key features of Strac MCP DLP

• Five tools — redact_text, detect_sensitive_data, detect_file, redact_file, detokenize.
• Redaction that does not leak — detections return data element types and positions, not the matched values, so redacting text never hands the same data back to the model. Opt in with include_matched_text.
• Files as well as text — images, PDFs, scans, configs and source code, using OCR and content-based classification. .env, .json, .yaml and id_rsa are read as text rather than images.
• Four redaction styles — [REDACTED], blank, XXXXXXX, or a vault token link an authorised person can resolve later.
• Never destroys your files — redact_file writes a copy, refuses to target the source, and will not replace an existing file unless you ask.
• Bring your own key — every call is authenticated to your Strac account.

Use cases of Strac MCP DLP

• Sanitise tool output — a support ticket, CRM record or log line — before it reaches a prompt, a summary or a downstream system.
• Scan a repo or config directory for leaked credentials: AWS keys, GitHub tokens, database connection strings.
• Redact W-2s, IDs, invoices and scanned documents before sharing them.
• Check whether a block of text is safe to send onward, without altering it.
• Keep PII, PHI and PCI out of agent workflows that would otherwise pass it to a model provider unfiltered.

Strac beyond this MCP server

This server protects data you hand it. The Strac platform protects data where it already lives.

Strac connects to 60+ SaaS, cloud and database systems — Slack, Google Workspace, Microsoft 365, Salesforce, Zendesk, Jira, Confluence, GitHub, Notion, Box, Dropbox, OneDrive, SharePoint, Snowflake, Databricks, BigQuery, Postgres, MongoDB, S3 and Azure Blob among them — and applies the same detection, redaction and remediation there: agentless, continuous, under your policies, with a full audit trail.

That matters most when those systems are reached over MCP. When an AI agent — Claude Code, Claude Desktop, Cursor, GitHub Copilot or OpenAI Codex — pulls a Salesforce record or a Google Drive file through an MCP connector, Strac redacts the sensitive data inline, before the agent ever receives it. Every MCP tool invocation across your connected platforms is recorded too: what was touched, which identity was behind the prompt, and where sensitive data moved.

That is the difference between asking an agent to redact and enforcing it whether or not it asks.

See https://www.strac.io/mcp-integrations.