Agent Guard
Add constitutional guardrails and loop detection so long-running coding agents stop repeating mistakes and stay within your rules.
Overview
Agent Guard is an MCP server for the Ship phase that enforces constitutional guardrails and detects agent loops during autonomous runs.
What is this MCP server?
- Constitutional guardrails you can wire into MCP-connected agents
- Loop detection to cut infinite retry/tool-call spirals
- Stdio MCP server published as npm package agent-guard-mcp v0.1.3
- Pairs with other mdfifty50-boop agent safety MCP servers in the same family
- Local stdio transport—no hosted API key in the server manifest
- Server version 0.1.3
- npm package identifier agent-guard-mcp
- stdio MCP transport
What problem does it solve?
Autonomous coding agents can ignore your rules and get stuck in expensive tool-call loops with no built-in stop condition.
Who is it for?
Solo builders running MCP-enabled agents on real repos who want lightweight policy and loop protection without building custom middleware.
Skip if: Teams that only need one-off chat completions with no tools, or anyone expecting full penetration testing or SIEM-style threat intel from this package alone.
What do I get? / Deliverables
After you register agent-guard-mcp, your agent client can apply guardrails and flag loops so runs fail safe instead of spiraling.
- Registered stdio MCP server exposing guardrail and loop-detection capabilities
- Agent sessions that can consult guard tools before unsafe repetition escalates
- Local npm-based install path (agent-guard-mcp 0.1.3)
Recommended MCP Servers
Journey fit
How it compares
Runtime guardrails MCP integration, not a static security audit skill or a session replay debugger.
Common Questions / FAQ
Who is Agent Guard for?
Indie and solo builders who deploy Claude Code, Cursor, Codex, or similar agents with MCP and need constitutional rules plus loop detection during autonomous work.
When should I use Agent Guard?
Use it when agents call tools repeatedly, edit production-adjacent code, or follow policies you want enforced before you trust longer autonomous sessions.
How do I add Agent Guard to my agent?
Install the npm package agent-guard-mcp (v0.1.3), add a stdio MCP server entry in your client config pointing at that package, and restart the agent so guardrail and loop tools are available.