AWS Security Tools, built in the open.
A curated catalogue of open-source security tooling contributed to aws-samples and beyond by the India Security Specialist Team. Each card opens the project's collateral — overview, architecture, capabilities and links.
AWS India Compliance MCP Server
Read-only assessment of AWS infrastructure against Indian regulatory frameworks — DPDP Act 2023, RBI IT Governance, SEBI CSCRF and CERT-In. 7 MCP tools; works with Kiro, Claude and Cursor.
AI Enclave
Securing GenAI & agentic workloads on AWS — "a deterministic boundary around a probabilistic system." An interactive walkthrough of the four objectives: secure WHERE, HOW, WHAT and WHEN.
Zero-Trust Agent Identity Governance
Continuous zero-trust governance for AI agent identities at organizational scale. Detects drift, enforces tagging, verifies permissions boundaries and trust policies, and monitors behaviour via CloudTrail Lake.
Next contribution
Placeholder for a further aws-samples security tool from the Admin team. Details and repository URL to be added.
Repositories & packages
AWS India Compliance MCP Server
Overview
An MCP (Model Context Protocol) server that evaluates AWS infrastructure against Indian regulatory frameworks. It performs read-only assessments and works with Kiro, Claude Desktop, Cursor and other MCP-compatible clients. Findings carry confidence ratings and evidence, but the tool does not constitute legal advice or compliance certification.
MCP tools
| Tool | What it does |
|---|---|
scan_aws_account | Discovers resources via AWS Config and assesses all frameworks; returns a compact (~2–3K token) summary with posture scores, gap counts and top critical findings. Supports tag filtering, exceptions, SEBI tiering and file saving. |
scan_control_tower | Enumerates enabled guardrails per OU, recommends missing ones per framework, with a domain-coverage breakdown. |
get_compliance_gaps | Paginated drill-down into gaps from the latest scan; filter by framework, risk level or domain (default 20/page, max 50). |
search_regulatory_text | Searches regulatory text from government sources, falling back to bundled mappings if sites are unreachable. |
list_control_domains | Lists framework domains: DPDP (10), RBI (7), SEBI (6), CERT-In (4). |
generate_conformance_pack | Produces deployable AWS Config conformance-pack YAML for a framework. |
format_report | Creates auditor-grade DOCX (default) or Markdown reports; can reuse cached scan results automatically. |
Maintainer tools (regulatory updates, mapping management) are CLI-only and not exposed via MCP.
Quick start
No install needed — uvx handles it. Add to your MCP config:
{
"mcpServers": {
"aws-india-compliance": {
"command": "uvx",
"args": ["aws-india-compliance@latest"],
"env": { "AWS_PROFILE": "my-sso-profile" }
}
}
}
Config locations: Kiro .kiro/settings/mcp.json, Claude Desktop claude_desktop_config.json, Claude Code ~/.claude/mcp.json. Requires uv (macOS: brew install uv).
Prerequisites
- Python 3.10+
- AWS Config recorder enabled
- Read-only IAM credentials (IAM Identity Center / SSO recommended)
- Config Aggregator for org-wide scans (auto-discovered)
Key features
- Confidence scoring — every gap rated High / Medium / Low with evidence, timestamps and rationale.
- CERT-In checks — incident pipeline (GuardDuty / EventBridge / SNS), 180-day log retention, NTP sync, Security Hub.
- Per-account breakdown for org-wide scans, each with its own posture score.
- SEBI entity tiering — MII (C-SOC + BYOK), qualified_re (BYOK), other_re (baseline).
- Nuanced data localization — storage outside India high-risk; compute medium-risk advisory; global services excluded.
- Exception management — auto-suppresses
AWSControlTowerExecutionroles; custom rules by pattern/tag. - Tag filtering — include (
filter_tags) or exclude (exclude_tags) resources. - Reports — DOCX with cover page, colour-coded scores and per-OU sections; or Markdown.
Control domains
DPDP (10)
Lawful Processing, Data Minimization, Privacy Notices, Data Principal Rights, Breach Notification, Security Safeguards, Retention Limits, Cross-Border Transfer, Children's Data, Significant Data Fiduciary Obligations.
RBI (7)
IT Governance, IT Infrastructure, IT Risk Management, Information Security, Cyber Security, BC/DR, IS Audit.
SEBI (6)
Cyber Governance, Risk Identification, Protection, Detection, Response, Recovery.
CERT-In (4)
Incident Reporting Readiness, Log Retention (180 days), NTP Sync, Reportable Incident Awareness.
Resource checks
S3 (encryption, Block Public Access, versioning, logging, Object Lock, TLS), RDS (encryption, public access, Multi-AZ, SSL), DynamoDB (KMS, TTL, PITR), Lambda (env secrets, DLQ), EC2 (public IP, IMDSv2, EBS encryption), EKS (secrets encryption, endpoint visibility, control-plane logging), ECS (Container Insights), CloudTrail (validation, KMS, CW Logs), KMS (rotation, BYOK), API Gateway / CloudFront (WAF), SQS / SNS (encryption), SageMaker (internet access, KMS, VPC), IAM roles (over-privileged policies), VPC Flow Logs, Security Groups (open SSH/RDP), Secrets Manager (rotation), AWS Backup (plans, Vault Lock), Inspector (enablement). RBI scans also flag resources outside ap-south-1 / ap-south-2.
Sample prompts
- "Scan my AWS account in ap-south-1 for DPDP and RBI compliance"
- "Scan my AWS organization for compliance"
- "Scan my AWS account as a SEBI MII entity"
- "Show me all critical DPDP gaps"
- "Scan my Control Tower and check guardrail coverage"
- "What does DPDP say about breach notification timelines?"
- "Format my last scan as a Word document"
- "Generate an AWS Config conformance pack for RBI"
AI Enclave: Securing GenAI & Agentic Workloads on AWS
Four objectives
- Secure WHERE — infrastructure & network boundary (VPC, PrivateLink, SageMaker isolation, Nitro Enclaves, IAM/SCP identity boundary, data localization).
- Secure HOW — design & development lifecycle (STRIDE threat modeling, PR scanning, SCA/SBOM, repo integration).
- Secure WHAT — deploy & release gate (pen testing, release testing, exploit validation, compliance gate).
- Secure WHEN — detect & respond at runtime (GuardDuty investigation agent, Security Hub, EventBridge auto-remediation, runtime guardrails).
Bank compliance mapping
Team work
AWS services in the enclave
Zero-Trust Agent Identity Governance
What it does
Treats every AI agent as an untrusted principal that must continuously prove its identity, justify its permissions
and have its behaviour monitored. It detects drift, enforces tagging compliance, verifies IAM permissions boundaries,
validates sourceIdentity trust policies, monitors behavioural drift via CloudTrail Lake, and verifies SCP enforcement — automatically.
Active governance checks
Runs twice daily.
| Check | Severity |
|---|---|
Tag compliance — IdentityType, AgentType, AgentScope | MEDIUM · auto |
| Permissions boundary attached to agent IAM roles | HIGH |
sourceIdentity enforcement in trust policies | HIGH |
| Behavioural monitoring (scope vs CloudTrail Lake) | HIGH |
SCP enforcement of aws:PrincipalTag/IdentityType | CRITICAL |
Real-time lifecycle governance
Event-driven.
| Event | Action |
|---|---|
CreateRole matching agent pattern | Auto-attaches standard permissions boundary |
AttachRolePolicy with unauthorized policy | Auto-detaches the policy |
Zero-trust principles
- No implicit trust — every request re-authenticates via short-lived credentials.
- Least privilege — permissions boundary + SCP + minimal identity policy.
- Continuous verification — actions logged, monitored, alerted.
- Assume breach — blast radius contained, attribution immutable.
- Micro-segmentation — unique identity per agent, layered authorization.
Architecture
Deploys across two AWS accounts — a management account for organization-level controls and a governance (child) account for all governance infrastructure.
┌─────────────────────────────────────────────────────────────┐
│ Management Account │
│ OrganizationStack: │
│ - SCPs denying actions without IdentityType tag │
│ - Cross-account EventBridge forwarding rules │
│ - Delegated Security Hub administrator config │
└──────────────────────────────┬──────────────────────────────┘
│
┌──────────────────────────────▼──────────────────────────────┐
│ Governance (Child) Account │
│ IdentityStack: AgentCore workload identities │
│ GovernanceIAMStack: IAM roles + permissions boundaries │
│ MonitoringStack: DynamoDB + Drift Agent + EventBridge │
│ + IAM Lifecycle Lambda + CloudTrail │
│ DashboardStack: CloudFront + Cognito + API Gateway │
│ + Lambda handlers (5 endpoint groups) │
│ AgentCoreStack: Strands SDK supervisor on Runtime │
└─────────────────────────────────────────────────────────────┘
Tech stack
Key services used
- AgentCore Identity — workload identity lifecycle (create, tag, list, delete).
- AgentCore Runtime — hosting governance agents with A2A protocol.
- AgentCore Policy — Cedar-based application-layer authorization.
- AWS IAM — permissions boundaries, SCPs, session tags for ABAC.
- CloudTrail Lake — behavioural monitoring via SQL on agent activity.
- Security Hub — centralized finding aggregation (dual-write with DynamoDB).
- EventBridge — real-time IAM lifecycle automation.
- DynamoDB — findings and orchestration-run storage.
Project structure
packages/
├── agents/ # Python - Strands SDK governance agents
│ └── src/ # Supervisor + sub-agents (drift, rotation, provisioning, decommission)
├── backend/ # TypeScript - Lambda handlers + services
│ └── src/
│ ├── handlers/ # Lambda entry points (drift-agent, identities, findings, agents, policies)
│ ├── models/ # Data models (drift findings, workload identities)
│ └── services/ # Check modules (permissions boundary, sourceIdentity, behavioral, SCP)
├── dashboard/ # TypeScript + React - Cloudscape UI
│ └── src/
│ ├── pages/ # Home, Identities, Findings, Agents, Policies
│ └── api/ # API client
└── infrastructure/ # TypeScript - CDK stacks
├── bin/ # CDK app entry point
└── lib/ # Stack definitions
Next contribution
Coming soon
This slot is reserved for an additional security tool from the Admin team's aws-samples contributions. Provide the repository URL and collateral and it will be filled in here — the card, detail page and GitHub links row are already wired up.