Agent Guardrail MCP Server

eren-solutions/agent-guardrail

This is a policy enforcement layer that sits between your AI agent framework and the actions it wants to take. Instead of filtering LLM outputs, it evaluates every tool call, shell command, or API request against allowlists, denylists, and spend caps before execution. You get a Python library with a PolicyEngine for LangChain, CrewAI, or custom agents, plus an optional HTTP proxy for multi-agent orchestrators. Every action gets logged to a SQLite flight recorder with full replay capability. Ships with three policy templates (restrictive, moderate, permissive) and includes automatic risk scoring. The hosted API mode adds metered billing via Bitcoin for agent-to-agent governance scenarios. Reach for this when you need runtime control over what agents can actually do, not just what they say.

tools: 6
transport: STDIO
last updated: Mar 14, 2026

View on GitHub

Tools

6 tools

Public tool metadata for what this MCP can expose to an agent.

evaluate_actionEvaluate whether an agent action is allowed by the guardrail policy. Call this BEFORE executing any tool, shell command, or HTTP request.5 params

Evaluate whether an agent action is allowed by the guardrail policy. Call this BEFORE executing any tool, shell command, or HTTP request.

Parameters* required

action_typestring

Category of action: tool_call, shell, http, file_read, file_write, database, api_call.

agent_idstring

Unique identifier of the agent. Must be registered first via register_agent.

cost_usdnumber

Estimated cost in USD for spend tracking and budget enforcement.default: 0

targetstring

Target resource path or URL, e.g. /etc/passwd, https://api.example.com.default:

tool_namestring

Name of the tool being invoked, e.g. bash, write_file, curl.default:

register_agentRegister a new agent with the guardrail system. Must be called before evaluate_action.3 params

Parameters* required

descriptionstring

Short description of what this agent does.default:

frameworkstring

Agent framework: langchain, autogen, crewai, claude-code.default:

namestring

Human-readable name for the agent, e.g. code-reviewer, data-analyst.

list_agentsList all agents currently registered in the guardrail system. Shows active and killed agents. Returns: Array of agent records with fields: id, name, framework, description, status, created_at.

List all agents currently registered in the guardrail system. Shows active and killed agents. Returns: Array of agent records with fields: id, name, framework, description, status, created_at.

No parameter schema in public metadata yet.

get_statsRetrieve aggregate statistics from the guardrail system including evaluation counts, policy data, and agent metrics. Returns: Statistics object with fields: total_evaluations, allowed_count, denied_count, agent_count, policy_count.

Retrieve aggregate statistics from the guardrail system including evaluation counts, policy data, and agent metrics. Returns: Statistics object with fields: total_evaluations, allowed_count, denied_count, agent_count, policy_count.

No parameter schema in public metadata yet.

kill_agentEmergency kill-switch: immediately block ALL actions for an agent. Use when an agent is misbehaving or compromised.1 params

Emergency kill-switch: immediately block ALL actions for an agent. Use when an agent is misbehaving or compromised.

Parameters* required

agent_idstring

Unique identifier of the agent to kill. All subsequent actions will be denied.

unkill_agentRe-enable a previously killed agent, restoring its ability to have actions evaluated.1 params

Re-enable a previously killed agent, restoring its ability to have actions evaluated.

Parameters* required

agent_idstring

Unique identifier of the agent to reactivate.

Readme

Agent Guardrail

Action-level governance for AI agents — control what they DO, not what they SAY.

The Problem

AI agents are getting tool access. They can run shell commands, make API calls, read files, spend money. But most "guardrails" only filter what agents say — not what they do.

Real incidents:

AutoGPT autonomously spent $10K+ on API calls in a single session
ChaosGPT attempted to access military systems and recruit other AI agents
Air Canada chatbot invented a refund policy that cost the airline $800+

You need action-level control. Not output filtering.

What Agent Guardrail Does

Agent Framework --> Agent Guardrail --> {allow, deny, require_approval}
                                    --> Flight Recorder logs everything

View the full README on GitHub

Details

Repository: github.com/eren-solutions/agent-guardrail
Website: github.com/eren-solutions/agent-guardrail
Category: AI & LLM Tools
Package: agent-guardrail
Transport: STDIO
Registry: MCP registry, active

Related AI & LLM Tools MCP servers

View all

Ragfieldcure/fieldcure-mcp-ragMCP RAG server with hybrid search, multi-KB support, and AI-powered chunk contextualization.0stars

Giskard Memorygiskard09/giskard-memoryPay-per-use semantic memory for AI agents. SHA256 commitment, Ed25519 sig, Lightning.0stars

User Promptjacob-j-thomas/user-context-retrieval-mcp-serverAn MCP server that enables AI agents to ask users clarifying questions via pop-up terminal windows.0stars

Memory Ftskurovu146/claude-memory-mcpLong-term memory for Claude Code with SQLite FTS5 search and BM25 ranking0stars

Agent Guardmdfifty50-boop/agent-guard-mcpConstitutional guardrails and loop detection for AI agents0stars

Agent Securitymdfifty50-boop/agent-security-mcpSecurity scanning and threat detection for AI agents0stars