# Pangea Security APIs MCP > Pangea Security APIs is an essential security layer for building LLM applications. It lets your AI client automatically scan inputs and outputs for sensitive data (PII), detect prompt injections, check IP origins against embargo lists, and manage user access rights before the information ever hits your model. ## Overview - **Category:** fort-knox - **Price:** Free - **Tags:** ai-guardrails, pii-redaction, threat-intelligence, audit-logging, prompt-injection, data-privacy ## Description When you build any application around a large language model, security is the biggest headache. You can’t trust what comes in or what goes out. This MCP gives you a unified way to enforce data rules right at the start of your process. It lets your agent automatically scrub plain text and structured JSON objects for private information; it also scans chat completions and prompts before they execute, stopping malicious inputs like prompt injections cold. If you're tracking activity, you get more than just a log file—you can search through all historical events using natural language queries. You can also validate user identity by starting sign-in flows or checking if an IP address is coming from a restricted region. Because it’s hosted on Vinkius, connecting this MCP to your workflow means you don't have to build custom middleware; you just connect and start securing everything. ## Tools ### ai_guard_prompt Analyzes and redacts malicious or sensitive content found in LLM prompts. ### ai_guard_text Scans any given text for PII, malicious patterns, and prompt injection attempts. ### aidr_chat_completions Guards entire LLM chat completions while logging and tracing every interaction securely. ### audit_log_bulk Creates multiple secure records of activity in your audit log at once. ### audit_log Records a single, specific security event into the tamper-proof audit trail. ### audit_search_results Retrieves pages of filtered search results from your historical audit log. ### audit_search Searches the entire audit history using natural language questions. ### authn_flow_complete Finalizes a user authentication flow and returns active session tokens. ### authn_flow_start Initiates the process for a user to sign up or log in. ### authn_flow_update Updates the state of an authentication flow, like submitting a password or OTP code. ### authn_session_list Retrieves a list of all currently active user sessions for management. ### authn_session_logout Invalidates and ends one or more existing user login sessions. ### authn_user_create Creates a new user account programmatically within your system. ### authz_check Determines if a specific user has permission to perform an action on a resource. ### authz_list_resources Lists all the resources that a given subject is authorized to access. ### authz_tuple_create Defines specific relationship rules for managing resource permissions (AuthZ). ### domain_whois Retrieves public WHOIS details associated with a given domain name. ### embargo_ip_check Checks if an IP address originates from any country that is currently under embargo. ### embargo_iso_check Verifies a two-letter ISO code against known lists of restricted countries. ### file_scan Scans an uploaded file to detect and flag any signs of malware or threats. ### intel_reputation Fetches reputation scores for specific domains, URLs, or file hashes. ### ip_geolocate Determines the physical location data associated with a given IP address. ### ip_proxy Checks if an incoming IP address is masking its true source because it's a proxy. ### ip_reputation Gets the overall reputation score and verdict for an entire IP address. ### ip_vpn Checks to see if an incoming IP address is masking its source because it's a VPN connection. ### password_breached Compares a password hash prefix against known databases of leaked credentials. ### redact_structured Removes specific fields from a JSON object based on defined paths (JSONPath). ### redact_text Scans and removes sensitive data from any block of plain text. ### sanitize_file Applies defined sanitization rules to a file, returning a cleaned version for safe use. ### share_delete Removes files or entire folders from the secure sharing area. ### share_folder_create Creates new organizational folders within the secure file sharing service. ### share_get Downloads a specific file or retrieves metadata about its properties. ### share_list Lists all objects within a designated bucket or folder, allowing for filtering. ### unredact Reverses the redaction process on data that was previously protected using FPE encryption. ### user_breached Checks if an email, phone number, or username appears in known public data breaches. ### vault_decrypt Performs cryptographic decryption on stored secret values. ### vault_encrypt Encrypts sensitive information to store it securely in the vault. ### vault_get Retrieves item details or specific secret values from your secure Vault storage. ### vault_key_generate Generates new symmetric or asymmetric cryptographic keys for use in the vault. ### vault_secret_store Stores a secret value or access token securely within the system's Vault. ## Prompt Examples **Prompt:** ``` Redact any PII from this text: 'My email is john.doe@example.com and I live in New York'. ``` **Response:** ``` I've processed the text using `redact_text`. The redacted version is: 'My email is and I live in '. ``` **Prompt:** ``` Search the audit logs for any 'login' events from the last 24 hours. ``` **Response:** ``` I've queried the audit logs using `audit_search`. I found 3 login events. Would you like to see the details for each entry? ``` **Prompt:** ``` Check if the IP address 1.1.1.1 is from an embargoed country. ``` **Response:** ``` Running `embargo_ip_check` for 1.1.1.1... The check returned that this IP is not associated with any embargoed countries. ``` ## Capabilities ### Data Privacy and Redaction Automatically find and scrub sensitive information from plain text or complex JSON objects. ### AI Input Guarding Analyze prompts and chat completions to detect malicious content, PII, and prompt injection attempts before processing. ### Security Auditing and Logging Maintain a chronological record of all actions taken in your system and search those records using plain language. ### Identity and Access Management Control who can access resources, manage user sessions, and programmatically create or update user accounts. ### Threat Intelligence Vetting Check IP addresses for geopolitical embargoes, detect proxies, and scan files for known malware signatures. ## Use Cases ### Handling Customer Support Chats A support agent needs to log a customer complaint that includes their account number and personal address. Instead of copying the whole thing into the system, your agent runs `ai_guard_text` first. It automatically strips all PII, so the security team gets a usable, non-sensitive report. ### Building Financial Agents A financial modeling bot needs to process quarterly reports that are structured JSON files containing salary data. The agent uses `redact_structured` to zero out all the sensitive salary fields before handing the data off for analysis. ### Onboarding New Users A new employee needs access to a specific shared drive. Instead of manually asking an admin, your agent first runs `authz_check` and verifies if the user's role has permission for that resource before granting access. ### Processing External APIs Your application receives data from a third-party API endpoint. Before processing it, your agent calls `embargo_ip_check` to ensure the incoming connection IP address is not from a prohibited country. ## Benefits - Stop data leaks before they start. Use `redact_text` or `redact_structured` to strip PII from user inputs, ensuring your model never sees private customer data. - Protect against bad actors with AI Guarding. Tools like `ai_guard_prompt` and `aidr_chat_completions` detect prompt injections—the biggest risk in LLM workflows. - Achieve compliance easily. The MCP lets you use `audit_search` to query audit logs using plain English, making compliance checks fast instead of manual report generation. - Control access at the source. You can manage user sessions using `authn_flow_start` or check permissions with `authz_check` before running any critical code path. - Vet external data sources. Before your agent uses an IP address, run `embargo_ip_check` to guarantee it's not coming from a restricted region. ## How It Works The bottom line is that it wraps your existing AI logic in layers of mandatory security checks. 1. First, subscribe to this MCP on the Vinkius Marketplace and input your Pangea Token and Domain credentials. 2. Next, configure your AI client—like Cursor or Claude—to route sensitive inputs and outputs through the security tools provided by this MCP. 3. Finally, when a user interacts with your application, the tools run in real time, checking for threats, redacting data, and logging the event before proceeding. ## Frequently Asked Questions **How do I use Pangea Security APIs MCP to redact JSON data?** You call `redact_structured` and pass the JSON object along with a list of fields you want removed. This method is better than generic redaction because it uses specific JSON paths. **Is Pangea Security APIs MCP good for checking suspicious IPs?** Yes. Use `ip_geolocate` to find where the IP comes from, and then run `embargo_ip_check` or `ip_reputation` to see if it's flagged as a proxy or coming from an embargoed nation. **What is the difference between audit_log and audit_log_bulk in Pangea Security APIs MCP?** `audit_log` creates one single record for a specific event. Use `audit_log_bulk` when you need to create many related records at once, like logging ten user sign-ins. **Can I use Pangea Security APIs MCP to check if my password is safe?** Yes, use the `password_breached` tool. It compares a hashed version of your password against public databases of leaked credentials to tell you if it's compromised. **Does Pangea Security APIs MCP help with general data storage?** No, this MCP handles security and access control, not storage. You use `vault_encrypt` or `vault_secret_store` to secure the *data*, but you need a separate service for actual file storage.