# Pangea MCP for AI Agents AI Agent Connect > Pangea MCP helps you put guardrails on your AI applications. It handles PII redaction, scans for prompt injections, checks IP reputations, and keeps a tamper-proof audit log. It's about making sure your AI doesn't leak secrets or get tricked by malicious inputs. ## Overview - **Category:** fort-knox - **Price:** Free - **Endpoint:** https://edge.vinkius.com/vk_preview_6PGcCGWSsRTH8ftWAASrBfIUfCx2NlDoXWkeA96h/ai-agent-connect - **Tags:** ai-guardrails, pii-redaction, threat-intelligence, audit-logging, prompt-injection, data-privacy ## Description Pangea puts a security gate in front of your AI applications. If you're building a tool that handles customer data, you know the nightmare of a potential leak. This Connector solves that by scrubbing sensitive info like social security numbers or private emails before they ever touch your model. It also acts as a shield against prompt injection, making sure users can't trick your agent into doing something it shouldn't. You get a searchable history of every action your agent takes, which is a lifesaver for compliance. It also handles the heavy lifting of threat intelligence, like checking if a user is hiding behind a VPN or coming from a blocked country. It even lets you manage user sessions and permissions directly through your agent's interface. By adding this to your setup through the Vinkius catalog, you're basically giving your AI a set of professional guardrails. It turns a scary security task into a background process that just works. You no longer have to worry about whether your model is being manipulated or if your logs are being tampered with. It even helps you manage secrets, so you never have to hardcode API keys or other sensitive credentials in your configuration again. It's about moving from a maybe this is safe mindset to a we know this is secure reality. ## Tools ### file_scan Scan a file for malware. This protects your system from malicious uploads. ### intel_reputation Retrieve reputation scores for domains, URLs, or file hashes. It helps you spot phishing attempts. ### ip_geolocate Retrieve location data for an IP. It tells you where your users are actually coming from. ### ip_proxy Detect if an IP is a proxy. Use this to identify users trying to hide their location. ### ip_reputation Get reputation score and verdict for an IP. This helps you block known malicious actors. ### ip_vpn Detect if an IP is a VPN. It helps you enforce location-based restrictions. ### password_breached Check if a password hash prefix appears in breach data. This helps you identify compromised credentials. ### share_list List objects in a bucket or folder with filtering. It lets you browse your protected storage. ### unredact Decrypt data previously redacted using FPE. Use this when you need to see the original data again. ### ai_guard_text Scans text for PII and malicious content. It stops bad data from reaching your model. ### aidr_chat_completions Guards LLM chat completions. It adds integrated logging and tracing to your chat flows. ### audit_log_bulk Creates multiple secure audit log entries at once. This is great for batch processing. ### audit_log Creates a single secure audit log entry. Use this to track individual actions. ### audit_search_results Paginate through audit search results. It helps you navigate through large volumes of logs. ### audit_search Search the audit log using natural language. You can ask questions about your security history. ### authn_flow_complete Finalize the auth flow and get session tokens. It's the final step for user sign-in. ### authn_flow_start Start a sign-up or sign-in flow. This initiates the authentication process for your users. ### authn_flow_update Update flow state like password submission. It handles the middle steps of a login. ### authn_session_list List active user sessions. Use this to see who is currently logged in. ### authn_session_logout Invalidate sessions. This lets you kick users out or end sessions. ### authn_user_create Programmatically create a user. It lets your agent set up new accounts automatically. ### authz_check Check if a subject has permission for an action. This is your primary tool for access control. ### authz_list_resources List all resources a subject can access. This helps with auditing user permissions. ### authz_tuple_create Define relationships for AuthZ. Use this to set up complex permission structures. ### domain_whois Get WHOIS details for a domain. It provides basic registration info for web addresses. ### embargo_ip_check Check if an IP originates from an embargoed country. This is vital for international compliance. ### embargo_iso_check Check a 2-character ISO country code against embargo lists. It's a quick way to verify geographic restrictions. ### redact_structured Redact specific fields in a JSON object using JSONPath. It's perfect for cleaning up API responses. ### redact_text Redact sensitive data from plain text. It replaces PII with placeholders automatically. ### sanitize_file Apply sanitization rules to a file and receive a cleaned version. It removes potentially harmful content from files. ### share_delete Delete files or folders. This is for managing your secure share storage. ### share_folder_create Create folders in Secure Share. Use this to organize your protected files. ### share_get Download a file or retrieve metadata. This is how you access files in your secure share. ### user_breached Check if an email, phone, or username appears in known data breaches. It helps you warn users about leaked info. ### vault_decrypt Perform cryptographic decryption. Use this to unlock secrets stored in your vault. ### vault_encrypt Perform cryptographic encryption. This is how you secure data before storing it in the vault. ### vault_get Retrieve item details or secret values from Vault. It's the standard way to fetch your API keys or secrets. ### vault_key_generate Generate symmetric or asymmetric keys in Vault. It helps you manage your own cryptographic keys. ### vault_secret_store Store a secret or token in Vault. Use this to save sensitive info securely. ### ai_guard_prompt Analyze and redact content in LLM inputs and outputs. It provides a safety layer for your prompts. ## Prompt Examples **Prompt:** ``` Can you check if my user's IP is coming from a blocked country? ``` **Response:** ``` Checking the IP 1.2.3.4... It is not currently on any embargo lists. You're good to proceed with the request. ``` **Prompt:** ``` I'm worried about privacy. Can you clean up this customer message before I send it to the database? ``` **Response:** ``` Certainly. I've processed the message for you. Here is the cleaned version: "My email is **** and my phone number is ****." ``` **Prompt:** ``` Show me a summary of all the security events from this morning. ``` **Response:** ``` I've searched the security logs for today. Here is what happened: | Time | Event | Status | | :--- | :--- | :--- | | 08:45 | User Login | Success | | 09:12 | File Access | Success | | 10:05 | Failed Login | Blocked | Would you like to see the details for any of these entries? ``` ## Capabilities ### Redact PII from text Automatically scrub sensitive information from plain text or JSON objects. ### Block prompt injections Detect and neutralize malicious instructions before they reach your model. ### Check IP reputations Identify suspicious traffic from proxies, VPNs, or embargoed countries. ### Create audit logs Maintain a tamper-proof record of every action your agent performs. ### Manage user sessions Handle authentication flows and session management directly through your agent. ### Scan files for malware Sanitize and scan files for threats before they enter your system. ## Use Cases ### PII Leak Prevention A developer building a customer support bot uses redact_text to ensure that if a customer types their credit card number, the AI only sees a redacted placeholder. ### Malicious Prompt Defense A company uses an AI agent to manage internal files. They use ai_guard_prompt to stop users from trying to trick the agent into deleting restricted folders. ### Compliance Auditing A compliance officer needs to know who accessed what. They use audit_search to ask Show me all login events from last week and get a clean list of results. ### Geo-Fencing A streaming service uses the Connector to check if a user is in a restricted country using embargo_ip_check before granting access to content. ## Benefits - No more manual PII scrubbing. Use redact_text and redact_structured to automatically clean data before it hits your logs. - Block prompt injections before they run. ai_guard_prompt and ai_guard_text act as a firewall for your LLM interactions. - Real-time threat detection. Use ip_reputation and ip_vpn to identify and block malicious actors or suspicious traffic. - Tamper-proof audit trails. audit_log and audit_search let you track every action your agent takes with natural language. - Automated identity management. authn_flow_start and authz_check let your agent handle user sign-ins and permissions safely. - Secure secret management. Use vault_get and vault_secret_store to handle API keys and credentials without hardcoding them. ## How It Works The bottom line is you get a production-ready security layer for your AI apps without writing custom backend code. 1. Get your Pangea Token and Domain from your Pangea console. 2. Add those credentials to your Connector client configuration. 3. Ask your agent to scrub data, check security status, or manage user sessions. ## Frequently Asked Questions **How does Pangea MCP help with data privacy?** It automatically scrubs sensitive info like emails and phone numbers from your AI's conversations so you don't accidentally leak private data. **Can I use Pangea MCP to block malicious users?** Yes, it checks IP reputations and detects VPNs or proxies, allowing your agent to block suspicious traffic before it reaches your app. **How do I audit what my AI agent is doing?** The Connector keeps a tamper-proof log of every action your agent performs. You can even ask your agent to search these logs using plain English. **Does Pangea MCP protect against prompt injection?** Yes, it analyzes inputs to detect and block attempts by users to trick your AI into performing unauthorized actions. **Is there a way to manage my secrets securely with Pangea?** You can use the vault features to store and retrieve API keys and other secrets without ever putting them in your code. **Can Pangea MCP handle complex user permissions?** It includes tools to manage user sessions and check if a specific user has permission to perform an action on a resource. **How can I protect my LLM from prompt injections or malicious content?** You can use the `ai_guard_prompt` tool to analyze LLM messages. It scans for prompt injections, PII, and other security risks, providing a safety score and redaction suggestions before the data is processed. **Can I search through my security logs using natural language?** Yes! The `audit_search` tool allows you to query your secure audit logs using natural language or structured queries, making it easy to find specific events without complex SQL. **How do I automatically redact PII from a block of text?** Use the `redact_text` tool. Simply provide the text, and Pangea will identify and mask sensitive information like emails, phone numbers, and names based on your configured rules.