Best for
- Enterprises needing strict governance for AI agents
- Teams meeting OWASP and regulatory compliance requirements
- Multi-agent systems requiring unified policy control
A
Agent Governance Toolkit MCP Server
Official
Policy enforcement, zero-trust identity, sandboxing, and reliability engineering for autonomous AI agents.
- Category
- Dev Tools #362 of 438
- Stars
- ★ 6.4k Very popular
- Transport
- stdio (local process)
- Runtime
- Node.js 18+
- Credentials
- No credential needed
- License
- MIT
- Last commit
39FMRS · D
This MCP server is officially maintained by Microsoft and offers robust governance for AI agents, ideal for scenarios requiring strict compliance and security. However, the manifest does not list specific tools, and installation depends on environment variables.
Strongest · Documentation 9/20 Weakest · Reliability 6/20
Why each score
- Reliability 6/20
- Evidence: The repository shows a structured MCP server path (server.json) indicating an implementation exists. However, I cannot access actual server code or tests. README mentions broad features (policy enforcement, audit) but specifics about the MCP server are thin. Missing unit tests or CI logs to verify init handshake or tool behavior. Thus reliability is low because there is no execution evidence.
- Security and permissions 8/20
- Evidence: The project emphasizes policy enforcement and zero-trust identity, suggesting a security-oriented design. Only env vars are policy mode and log level, non-secret. No credentials or secrets appear in install examples. Possible network access (tools like web search) but governed by policy. No evidence of dangerous operations but no explicit confirmation mechanisms documented. No red flags detected, but inability to inspect code for full least-privilege verification causes slight deduction.
- Maintenance 9/20
- Evidence: Repo is highly active with thousands of stars, frequent releases (v1.0.1), strong CI. Security policy and badges (OpenSSF Scorecard) indicate good practices. License present (MIT). Organization (Microsoft) provides clear ownership. Missing direct evidence of issue response time, but with organizational backing and clear maintainers, maintenance appears strong.
- Documentation 9/20
- Evidence: README is extensive, covering many languages, frameworks, and compliance, but not specific to the MCP server. Lacks detailed docs on MCP server tools, parameters, and limitations. No troubleshooting guide for this specific server. Overall project docs are rich, but MCP-server-specific documentation is insufficient, deducting points.
- Setup experience 7/20
- Evidence: README provides installation instructions for the specific MCP server package via npm and env vars. However, no detailed client configuration examples (e.g., Claude Desktop or VS Code). Setup steps are unclear, and troubleshooting is missing. Being static, I cannot verify installation smoothness. Hence moderate setup score.
Static review · not runListed 2026-08-07
Read the FMRS scoring method →Fit and risk
What it can accessRuns commands or code
Not for
- Simple chatbots without tool calls or autonomy
- Prototyping that requires unconstrained flexibility
- Performance-sensitive scenarios that cannot tolerate enforcement overhead
Required permissions
- Requires access to policy files and configuration
- May log tool call audit trails
- May access identity services for agent verification
Risks and side effects
- Misconfigured policies may block legitimate operations
- Over-restriction may hinder agent capabilities
- Audit logs may contain sensitive operation information
Setup
Before you start
Runtime：Node.js 18+
Other optional settings (2)
AGENTOS_POLICY_MODE
Policy enforcement mode: strict (block all violations) or permissive (warn only); optional.
AGENTOS_LOG_LEVEL
Log level: debug, info, warn, or error; optional.
The server is available as an npm package named 'agentos-mcp-server'. Install with npm install agentos-mcp-server. Configure environment variables AGENTOS_POLICY_MODE (strict or permissive) and AGENTOS_LOG_LEVEL (debug, info, warn, or error) to control policy enforcement and logging.
Check that it works
Confirm the agentos MCP server (npm package agentos-mcp-server, version 1.0.1) starts over stdio in your client and can perform policy-compliance checks; set AGENTOS_LOG_LEVEL=debug to verify the connection in logs.
Troubleshooting
- Check AGENTOS_POLICY_MODE configuration
- Validate policy file syntax and rules
- Review log level settings to ensure sufficient output
Things to try
Once connected, you can ask your AI assistant things like:
- Check whether my agent's actions comply with the governance policy in policy.yaml
- Set the policy mode to strict and evaluate what happens on a violating tool call
- Write a governance rule that blocks destructive operations like drop and delete
- Audit the decision records from this agent session and explain each allow or deny
Use cases
Enforce security policies when deploying AI agents
Audit tool calls for compliance
Implement zero-trust identity and sandboxing
Overview
The Agent Governance Toolkit MCP server enables building and managing policy-compliant AI agents with safety enforcement and compliance checking. Based on Microsoft's Agent Governance Toolkit, it provides policy enforcement, zero-trust identity, execution sandboxing, and reliability engineering capabilities, covering all of the OWASP Agentic Top 10. The server integrates via MCP, allowing client tools to enforce policies on agent actions, ensuring operations comply with organizational requirements.
Similar servers
Context7
80 · B
Upstash's official server providing up-to-date third-party library docs for AI coding assistants
NPM Sentinel MCP Server
79 · B
AI-powered NPM package analysis MCP server
MartinLoop MCP Server
79 · B
An independent control layer that gives AI coding agents budgets, brakes, and receipts
GitHub MCP Server
76 · B
GitHub's official MCP server for managing repos, issues, PRs, and workflows via natural language
Source revision c767f8333f41 Data synced 2026-10-11 Read the FMRS scoring method