SaferAgenticAI MCP — AI Safety Framework¶
What It Is¶
SaferAgenticAI MCP (NellInc/SaferAgenticAI) brings the Safer Agentic AI safety framework to coding assistants over the Model Context Protocol. It enforces governance guardrails, safety checks, and policy constraints on agent actions — ensuring agentic workflows comply with organizational safety policies before executing tool calls, file operations, or external API requests.
Tools Available¶
| Tool | Description |
|---|---|
| Safety check | Pre-execution validation of agent actions against policy |
| Policy enforcement | Rule-based guardrails for tool calls and code execution |
| Audit logging | Record of all safety decisions and policy violations |
Quick Start¶
uvx saferagenticai-mcp
Business Use Cases¶
- Production agent guardrails: Deploy Hermes agents in production with safety checks on file writes, network calls, and tool invocations
- Compliance enforcement: Ensure agents never access PII, write to restricted directories, or call unapproved APIs
- Multi-agent governance: Apply consistent safety policies across all agent instances in a team
- Audit trail: Every safety decision logged for SOC 2, HIPAA, or internal compliance review
Limitations¶
- Research-stage: Part of the NellInc research repository — not yet production-hardened
- Coding assistant focus: Primarily designed for coding agents (Codex, Claude Code); may need adaptation for business operations agents
- No remote transport: Local-only via
uvx