Add eren-solutions/agent-guardrail -- action-level governance for AI agents

[JarvisOnM4]

Server submission: agent-guardrail

Please add agent-guardrail to the Smithery registry.

Package details:

• PyPI package: agent-guardrail
• Version: 0.1.2
• GitHub: https://github.com/eren-solutions/agent-guardrail
• Hosted API: https://157-230-82-223.sslip.io/guardrail/health
• smithery.yaml: https://github.com/eren-solutions/agent-guardrail/blob/main/smithery.yaml

What it does:

Action-level governance for AI agents — controls what agents do, not just what they say.

Key features:

• Policy engine: allowlists, denylists, pattern matching on tool calls/targets
• Spend caps: per-agent USD limits with automatic enforcement
• Kill switch: emergency halt for any registered agent
• Flight recorder: full audit trail with session replay
• Approval gates: human-in-the-loop for high-risk actions
• Free tier: 100 evaluations/day per agent, no signup required

Already listed:

• MCP Registry: https://registry.modelcontextprotocol.io/v0.1/servers?search=agent-guardrail (active since 2026-03-14)

MCP tools provided:

• evaluate_action — evaluate an agent action against policies (allow/deny/require_approval)
• register_agent — register a new agent, get API key
• agent_stats — aggregate stats across all agents

Usage (hosted):

{
  "mcpServers": {
    "agent-guardrail": {
      "command": "python3",
      "args": ["mcp_guardrail_http.py", "--transport", "stdio"],
      "env": {
        "GUARDRAIL_ENDPOINT": "https://157-230-82-223.sslip.io/guardrail"
      }
    }
  }
}

Usage (self-hosted):

pip install agent-guardrail[proxy]
guardrail-proxy --port 8300