Safeprompt MCP server
{
"mcpServers": {
"safeprompt": {
"command": "npx",
"args": [
"-y",
"@safeprompt.dev/mcp@latest"
],
"env": {
"SAFEPROMPT_API_KEY": "<safeprompt_api_key>",
"YOUR_API_KEY": "<your_api_key>"
}
}
}
}{
"mcpServers": {
"safeprompt": {
"command": "npx",
"args": [
"-y",
"@safeprompt.dev/mcp@latest"
],
"env": {
"SAFEPROMPT_API_KEY": "<safeprompt_api_key>",
"YOUR_API_KEY": "<your_api_key>"
}
}
}
}{
"mcpServers": {
"safeprompt": {
"command": "npx",
"args": [
"-y",
"@safeprompt.dev/mcp@latest"
],
"env": {
"SAFEPROMPT_API_KEY": "<safeprompt_api_key>",
"YOUR_API_KEY": "<your_api_key>"
}
}
}
}{
"servers": {
"safeprompt": {
"command": "npx",
"args": [
"-y",
"@safeprompt.dev/mcp@latest"
],
"env": {
"SAFEPROMPT_API_KEY": "${input:safeprompt_api_key}",
"YOUR_API_KEY": "${input:your_api_key}"
}
}
},
"inputs": [
{
"id": "safeprompt_api_key",
"type": "promptString",
"description": "Safeprompt api key",
"password": true
},
{
"id": "your_api_key",
"type": "promptString",
"description": "Your api key",
"password": true
}
]
}{
"mcpServers": {
"safeprompt": {
"command": "npx",
"args": [
"-y",
"@safeprompt.dev/mcp@latest"
],
"env": {
"SAFEPROMPT_API_KEY": "<safeprompt_api_key>",
"YOUR_API_KEY": "<your_api_key>"
}
}
}
}file .mcp.json (in your project root)file claude_desktop_config.jsonfile ~/.cursor/mcp.jsonfile .vscode/mcp.jsonfile ~/.codeium/windsurf/mcp_config.jsoninstall npx -y @safeprompt.dev/mcpversion 0.1.0
What it does
SafePrompt is a prompt injection detection API that screens untrusted text before it reaches an LLM. It identifies prompt injection, jailbreaks, code injection patterns, and exfiltration attempts. It is for developers building chatbots, AI apps, and automations who want to reduce integration-boundary security risks.
- Check user input for prompt injection attacks before passing it to an LLM in a chatbot or support system.
- Detect jailbreak attempts and system-prompt extraction attempts in multi-turn conversations.
- Screen for code-injection patterns (XSS, SQL injection, template injection, command injection) embedded in user-supplied text.
Tools this server exposes
| Tool | Does |
|---|---|
| check | Analyzes text for prompt injection, jailbreaks, code injection, and exfiltration patterns; returns a safe boolean and list of detected threats. |
Before you enable it
An MCP server runs with whatever access you give it. Start with a development account rather than production, and check which tools the server exposes before you let an agent call them unattended.