Enkrypt AI MCP Server
Official# Enkrypt AI MCP Server
The Enkrypt AI MCP Server allows you to integrate red-teaming, prompt auditing, and AI safety analysis directly into any [Model Context Protocol (MCP)](https://modelcontextprotocol.io/introduction)–compatible client such as Claude Desktop or Cursor IDE.
With this server, you can analyze prompts, detect jailbreak attempts, simulate adversarial attacks, and bring AI safety tooling directly into your assistant-driven workflows.
---
## 🚀 Features
- Real-time prompt risk analysis
- Red-teaming via adversarial prompt generation
- Tool-based LLM monitoring using the MCP standard
- Seamless integration with Claude Desktop, Cursor IDE, and other MCP clients
---
## 💠 Installation
Before getting started, ensure you have [`uv`](https://docs.astral.sh/uv/getting-started/installation/) installed on your machine.
### 1. Clone the repository
```bash
git clone https://github.com/enkryptai/enkryptai-mcp-server.git
cd enkryptai-mcp-server
```
### 2. Install dependencies
```bash
uv pip install -e .
```
---
## 🔑 Get Your API Key
To use the Enkrypt tools, you’ll need a free API key from:
[https://app.enkryptai.com/settings/api](https://app.enkryptai.com/settings/api)
---
## ⚙️ Configuration
You can connect this MCP server to any MCP-compatible client. Here's how to do it with **Cursor** and **Claude Desktop**.
---
### 🖥️ Cursor
1. Open **Settings** → **MCP** tab in Cursor
2. Click **"Add new global MCP server"**
3. Paste the following config into the `mcp.json` file:
```json
{
"mcpServers": {
"EnkryptAI-MCP": {
"command": "uv",
"args": [
"--directory",
"PATH/TO/enkryptai-mcp-server",
"run",
"src/mcp_server.py"
],
"env": {
"ENKRYPTAI_API_KEY": "YOUR ENKRYPTAI API KEY"
}
}
}
}
```
Replace:
- `PATH/TO/enkryptai-mcp-server` with the absolute path to the cloned repo
- `YOUR ENKRYPTAI API KEY` with your API key
The server will launch and appear in your MCP tools list.
---
### 💬 Claude Desktop
1. Open the **Claude** menu in your system menu bar (not inside the app window)
2. Go to **Settings…** → **Developer** tab
3. Click **Edit Config**
This opens or creates the MCP config file at:
- macOS: `~/Library/Application Support/Claude/claude_desktop_config.json`
- Windows: `%APPDATA%\Claude\claude_desktop_config.json`
Replace the contents with:
```json
{
"mcpServers": {
"EnkryptAI-MCP": {
"command": "uv",
"args": [
"--directory",
"PATH/TO/enkryptai-mcp-server",
"run",
"src/mcp_server.py"
],
"env": {
"ENKRYPTAI_API_KEY": "YOUR ENKRYPTAI API KEY"
}
}
}
}
```
Make sure to:
- Set the correct repo path
- Paste in your API key
Finally, restart Claude Desktop. Once it reloads, you’ll see a hammer icon in the chat box, indicating your MCP tools are active.
TDQS
Scored across 28 tools
Multiple tools have overlapping purposes that could cause confusion. For example, add_redteam_task, add_agent_redteam_task, and add_custom_redteam_task all create red team tasks with subtle differences that may not be clear from their names alone. Similarly, add_model and add_model_from_url both add models but with different configurations, and guardrails_detect vs use_policy_to_detect both perform detection but with different approaches. The descriptions help, but the boundaries are unclear.
Most tools follow a consistent verb_noun pattern (e.g., add_model, get_model_details, list_models, remove_model), which is clear and predictable. However, there are minor deviations like guardrails_detect (noun_verb) and use_policy_to_detect (verb_noun_preposition_noun), which break the pattern slightly but are still readable. Overall, the naming is mostly consistent.
With 28 tools, the count feels excessive for the apparent scope of AI model and red teaming management. Many tools could be consolidated (e.g., multiple red team task additions, redundant detection tools), leading to a bloated interface. A well-scoped server for this domain would typically have 10-15 tools, making this set heavy and potentially overwhelming for agents.
The tool set covers a broad range of operations for managing AI models, red teaming tasks, deployments, and guardrails policies, including CRUD actions and workflow steps like hardening system prompts. However, there are minor gaps, such as no tools for updating or deleting red team tasks directly, and some operations rely on combinations of tools that might be less intuitive. Overall, the surface is largely complete for the domain.