# Aporia MCP for AI Agents AI Agent Connect

> Aporia lets you put guardrails on your LLMs and monitor their performance in real time. You can check for toxicity, PII leaks, and data drift directly through your AI client. It connects your observability data to your agent so you can audit dashboards and trigger monitors without leaving your chat interface.

## Overview
- **Category:** fort-knox
- **Price:** Free
- **Endpoint:** https://edge.vinkius.com/vk_preview_xK4X7s8PmKOy9iRpjDUr4oudy86jlYzJmpsm9w0q/ai-agent-connect
- **Tags:** llm-guardrails, model-monitoring, ai-safety, ml-ops, data-integrity, toxicity-detection

## Description

You're trying to get an LLM into production, but you're worried about it hallucinating, leaking private data, or just going off the rails. This connection gives you a way to check those concerns on the fly. Instead of jumping between your monitoring dashboard and your development environment, you can just ask your agent to check if a specific prompt triggered a violation or to pull the latest drift metrics. It brings the heavy lifting of MLOps into your daily workflow. You'll find it's a lot easier to stay on top of model health when you can query your custom dashboards and trigger active monitors with a simple sentence. It's a core part of the Vinkius ecosystem, making it easy to manage complex AI infrastructure without the usual context switching. You get a clear view of how your models are actually behaving in the wild, letting you catch issues before they become actual problems for your users.

## Tools

### list_dashboards
Get a list of all the custom observability dashboards you've built in your workspace.

### list_models
See every machine learning and LLM model currently tracked in your Aporia workspace.

### get_model
Pull out the specific architectural details and metadata for a single monitored model.

### list_monitors
View the specific monitors you've set up for a particular model to see what's being watched.

### trigger_monitor
Run a specific monitor immediately to check for data integrity or performance issues right now.

### validate_guardrails
Send a batch of messages to check for toxicity, PII leaks, or off-topic content against your rules.

### get_metrics
Fetch real-time performance and data drift numbers for any model you're tracking.

## Prompt Examples

**Prompt:** 
```
Which models are we currently tracking?
```

**Response:** 
```
I found 3 monitored models in your workspace:

| Model Name | ID |
| :--- | :--- |
| **Customer Churn Predictor** | `mdl_a1b2` |
| **GPT-4 Support Bot** | `mdl_x5y6` |
| **Recommendation Engine** | `mdl_j7k8` |

Would you like to see the monitors for any of these?
```

**Prompt:** 
```
Does this message violate any safety rules?
```

**Response:** 
```
I've checked that message against your safety guardrails. 

**Result:** ❌ Violation Detected
**Reason:** Prompt Injection

The engine detected malicious command structures meant to override the context. It was blocked safely. Should I pull the latest metrics to see if this is a recurring issue?
```

**Prompt:** 
```
What's the drift looking like for the churn predictor?
```

**Response:** 
```
Pulling the latest metrics for **Customer Churn Predictor** (`mdl_a1b2`):

*   **Inference Count:** Up 15% this week
*   **Accuracy:** Stable
*   **⚠️ Data Drift Warning:** Detected in the `user_tenure` feature.

Would you like me to list the monitors associated with that feature?
```

## Capabilities

### Block PII and toxic content
Validate messages against your safety rules to catch leaks and off-topic responses.

### Track model health
Pull real-time drift and performance metrics for any model in your workspace.

### Run on-demand checks
Trigger specific monitors instantly to verify data integrity or model behavior.

### Audit custom dashboards
Access and view your aggregated observability metrics through a natural conversation.

### Inventory your models
List all monitored machine learning models and fetch their specific architectural details.

## Use Cases

### Pre-deployment safety check
An AI engineer sends a batch of test prompts to validate_guardrails to ensure no PII is leaked before pushing to production.

### Sudden accuracy drop investigation
An MLOps engineer asks their agent to get_metrics for the churn predictor to see if a specific feature is drifting.

### Real-time toxicity auditing
A risk officer asks the agent to list_dashboards to review the latest toxicity trends across all customer-facing bots.

### On-demand integrity verification
A data scientist uses trigger_monitor to run a specific check on the recommendation engine's data integrity after a database update.

## Benefits

- Stop PII leaks before they happen by using validate_guardrails to check every prompt against your safety rules.
- Spot data drift faster by pulling live stats with get_metrics instead of refreshing your browser tabs.
- Run manual checks on demand using trigger_monitor to see if a model is still behaving as expected.
- See your entire fleet at once with list_models to keep your model inventory organized and accessible.
- Access your custom views instantly with list_dashboards to see the big picture of your AI infrastructure.
- Get architectural clarity by using get_model to pull specific details for any monitored system.

## How It Works

The bottom line is you get a direct line to your model observability data inside your favorite chat interface.

1. Subscribe to the Aporia MCP via the Vinkius catalog.
2. Enter your Aporia API key into your AI client settings.
3. Ask your agent to check guardrails, pull metrics, or view your custom dashboards.

## Frequently Asked Questions

**How does Aporia MCP help with LLM safety?**
It connects your safety rules directly to your AI client. You can ask your agent to check prompts for toxicity or PII leaks on the fly, ensuring your model stays within your defined guardrails.

**Can I use Aporia MCP to detect PII?**
Yes. The Connector allows your agent to run validation checks on messages to identify and block sensitive information like PII before it's processed or stored.

**How do I check for model drift with Aporia MCP?**
You can simply ask your agent to pull the latest metrics for any model in your workspace. It will return real-time data on performance and identify specific features where drift is occurring.

**Can I trigger a manual check with Aporia MCP?**
Absolutely. You can tell your agent to run a specific monitor immediately. This is great for verifying data integrity after a system update or a manual data injection.

**Is Aporia MCP good for MLOps teams?**
It's designed for MLOps. It streamlines how teams monitor model health, manage model inventories, and audit custom dashboards without needing to switch between different monitoring tools.

**How does Aporia MCP handle toxicity?**
It uses your configured guardrails to scan messages for hateful content or off-topic responses, providing an instant pass/fail check directly in your chat interface.

**Can my AI agent check for PII vulnerabilities in my conversation logs?**
Yes. Provide a message context and prompt the agent to validate guardrails for a specific Aporia project. It will query the Aporia rules engine and return actionable insights immediately — ensuring that your outputs remain safe and compliant before deployment.

**How quickly can I review data drift alerts triggered over the weekend?**
In seconds. During your Monday stand-up, ask your AI agent to fetch metrics for your primary model and list its monitors. The agent will pull up statistical variations and active alerts directly into your chat, saving you from navigating complex visual dashboards when you need a quick situational report.

**If a monitor seems stuck, can the agent force its execution?**
Absolutely. If you suspect an anomaly, ask the agent to manually trigger the specific monitor ID. Instead of scheduling a redundant pipeline, the agent hits the Aporia backend to force a run, ensuring your MLOps workflow isn't blocked by cadence limitations.