# SafePrompt (safeprompt.dev)

Detects prompt injection in text an app sends to an AI model, returning a safe/unsafe verdict via a single API call.

- Atlas card: https://inite.ai/en/atlas/safeprompt.dev
- Site: https://safeprompt.dev
- For: Developers and teams building LLM apps and agents that read user messages and retrieved documents
- Categories: security
- Use 24/100 · alive 82/100

## Used for

- screen user messages before a chatbot acts on them
- check retrieved documents before an agent reads them
- flag injected instructions in tool output before use

## Price

- from 29 USD/month, free plan (read 2026-10-29 on https://safeprompt.dev/pricing)

## For developers

- API docs: https://docs.safeprompt.dev/api-reference
- MCP server: dev.safeprompt/mcp (namespace; open on 2026-10-29, 2 tools)
- Read 2026-10-29

## MCP server, measured daily

Server: https://mcp.safeprompt.dev/mcp, checked every day since 2026-10-29 (5 checks).

- Answered: measuring, 5 daily checks so far
- Median first answer: 74 ms
- Sign-in: none
- Context its tool definitions take: about 443 tokens (2 tools)
- Tool definitions: A, 100/100
- Method: https://inite.ai/en/atlas/mcp/report

## Your data

- Certifications stated: gdpr, ccpa
- Training on your data: no (https://safeprompt.dev/privacy)
- Read 2026-10-29

---

Source: INITE Atlas, https://inite.ai/en/atlas/safeprompt.dev. JSON: https://inite.ai/api/atlas/safeprompt.dev. MCP: https://inite.ai/api/mcp (atlas_lookup).