Free Download

Claude Code Security Config for GCC High

630-line CLAUDE.md template. Drop it in your repo root and Claude Code inherits CMMC-aligned behavior automatically.

No spam. Relevant GCC High and CMMC content only.

NSA AISC May 2026 OWASP Agentic AI CMMC L2 · NIST 800-171 MITRE ATLAS

Claude Code Security Configuration for GCC High

A security-hardened CLAUDE.md that governs how Claude Code behaves inside a Microsoft GCC High environment — CUI boundary rules, endpoint enforcement, MCP security, and CMMC control mapping in one file.

What This Is

One File. Claude Code Inherits CMMC-Aligned Behavior.

When Claude Code finds a CLAUDE.md at the repository root, it reads and follows those instructions for the entire session. This configuration turns a general-purpose AI coding assistant into one that understands your GCC High boundary, refuses to write code that violates it, and flags DFARS-reportable events when it spots them.

It covers 14 sections — from CUI boundary enforcement and hardcoded-secret prohibition to MCP server security and GitHub Enterprise Server integration. References the NSA AISC Cybersecurity Information Sheet on MCP security published May 2026 and aligns with the new OWASP Agentic AI Top 10.

Drop it in your repo root. That's it.

Discuss Custom Config

630 lines · UNCLASSIFIED // FOUO · CMMC L2 · NIST 800-171 R2 · DFARS 252.204-7012

14 Sections Covered
01Environment Identity & GCC High Boundary
02Critical Security Constraints (CUI, Secrets, Auth)
03GCC High Endpoint Reference (complete table)
04MCP Server Security (NSA AISC May 2026)
05Code Generation Standards
06CMMC Control Mapping in Code Headers
07AI Agent Security (OWASP Agentic AI Top 10)
08Prohibited Actions (13 hard rules)
09Required Patterns Checklist
10Dependency & Supply Chain Security
11GitHub Enterprise Server Integration
12Evidence & Immutability (WORM storage patterns)
13Incident & DFARS Reportable Event Detection
14Review Gates & CISO Sign-Off Requirements
Preview — Section 3

GCC High Endpoint Quick Reference

Every API call Claude Code generates uses these endpoints. Commercial .com endpoints fail silently or route data outside your authorization boundary. This table is embedded in the config so Claude Code never gets it wrong.

QUICK REFERENCE — GCC High vs Commercial
What you wantCommercial ✗ WRONGGCC High ✓ CORRECT
Login / Authlogin.microsoftonline.comlogin.microsoftonline.us
Microsoft Graphgraph.microsoft.comgraph.microsoft.us
Azure Resource Managermanagement.azure.commanagement.usgovcloudapi.net
Key Vault.vault.azure.net.vault.usgovcloudapi.net
Blob Storage.blob.core.windows.net.blob.core.usgovcloudapi.net
Log Analyticsapi.loganalytics.ioapi.loganalytics.us
Sentinel Ingest.ods.opinsights.azure.com.ods.opinsights.azure.us
MDE APIapi.securitycenter.microsoft.comapi.securitycenter.microsoft.us
Azure OpenAI.openai.azure.com.openai.azure.us
Cosmos DB.documents.azure.com:443.documents.azure.us:443
Intune / MDMmanage.microsoft.commanage.microsoft.us
PowerShell env(default / AzureCloud)AzureUSGovernment
Graph PowerShell(default)-Environment USGov
Exchange Online(default)O365USGovGCCHigh

The full config embeds this table plus authentication code patterns in Python, PowerShell, and TypeScript so Claude Code generates the correct endpoint in every language.

Preview — Section 8

13 Things Claude Code Must Never Do in GCC High

These prohibited patterns are embedded in the config. Claude Code refuses to generate them and explains why — keeping your developers from accidentally violating CMMC controls or routing CUI outside the boundary.

✗Write code that sends data to commercial .com Microsoft endpoints
✗Hardcode any credential, token, key, or connection string
✗Log or print authentication tokens, API keys, or secrets
✗Generate code that disables TLS verification (verify=False, InsecureSkipVerify)
✗Create outbound connections to non-allowlisted external domains
✗Write code that reads or copies files from CUI-designated paths
✗Generate self-signed certificate creation for production use
✗Create code that stores audit logs in non-immutable locations
+ 7 more in the full config
Preview — Section 2

No Secrets. Ever.

The config enforces Key Vault patterns and flags any generated code that puts a secret in plain text — even in a comment.

# PROHIBITED — Claude Code will refuse to generate api_key = "sk-..." client_secret = "abc123" connection_string = "Server=...;Password=..." # CORRECT — what the config enforces from azure.keyvault.secrets import SecretClient from azure.identity import DefaultAzureCredential credential = DefaultAzureCredential() client = SecretClient( vault_url="https://VAULT.vault.usgovcloudapi.net", credential=credential ) secret = client.get_secret("my-secret-name")
Preview — Section 4

MCP Security — NSA Guidance Applied

The May 2026 NSA AISC advisory on MCP security is embedded directly. Claude Code enforces authorized-only MCP server lists, input sanitization before every MCP call, and audit logging of all tool invocations.

// PROHIBITED — dynamic tool invocation const tool = userInput.requestedTool; // NEVER await mcpClient.call(tool, params); // CORRECT — explicit allowlist const ALLOWED = ['search_knowledge_base', 'get_doc_status']; if (!ALLOWED.includes(requestedTool)) { throw new Error('Tool not in authorized list'); }
Free Download

Get the Complete 630-Line Configuration

Covers all 14 sections including CMMC control mapping patterns, WORM evidence storage, DFARS incident detection, GitHub Enterprise Server integration, and the full NSA MCP security framework applied to Claude Code in GCC High.

Get a Custom Version

Fulcrum Advisory also builds custom CLAUDE.md configurations tailored to your specific GCC High tenant, CMMC scope, and development stack.

Why This Exists

AI Coding Assistants Are Coming to GovCon Whether You're Ready or Not

Defense contractors are adopting AI coding tools — GitHub Copilot, Claude Code, Cursor — faster than their security teams can assess them. The risk isn't that the tools are malicious. The risk is that they're trained on commercial patterns and will confidently generate code that routes data to graph.microsoft.com instead of graph.microsoft.us, stores a connection string in plain text, or calls a commercial Azure OpenAI endpoint from inside your GCC High boundary.

A CLAUDE.md at the repo root changes the behavior before any code is written. It's the equivalent of a new developer reading your security policy on day one — except Claude Code actually follows it, every session, without exception.

This configuration was built from experience deploying production AI systems inside a CMMC-scoped GCC High environment, combined with the NSA's May 2026 guidance on MCP server security and the new OWASP Agentic AI Top 10. It's not a theoretical framework — it's what we use.

AI in GCC High Services GCC High Parity Map Regulatory Watch

Need a Custom Configuration?

We build CLAUDE.md configurations tailored to your specific GCC High tenant, CMMC scope, and development stack.

Schedule a Call