AWS Security Agent MCP Server

An AWS Labs Model Context Protocol (MCP) server for AWS Security Agent — automated security scanning and penetration testing.

This MCP server provides full access to the AWS Security Agent service, enabling developers to scan source code for vulnerabilities, run penetration tests against live applications, manage integrations, and apply auto-generated fixes — all from any MCP-compatible client.

Features

• Code security scanning — zip, upload, scan source code, get findings with fixes
• Penetration testing — test live applications via target domains
• Full API access — call_api tool exposes all SecurityAgent operations
• Auto-provisioning — creates agent space and IAM service role on first use
• Respects .gitignore — excludes ignored files from packaging

Prerequisites

1. uv installed
2. Python 3.10+ (uv python install 3.10)
3. AWS credentials configured (via aws configure, SSO, or environment variables)

Installation

Using uvx (recommended)

{
  "mcpServers": {
    "awslabs.security-agent-mcp-server": {
      "command": "uvx",
      "args": ["awslabs.security-agent-mcp-server@latest"],
      "env": {
        "AWS_PROFILE": "default",
        "AWS_REGION": "us-east-1",
        "FASTMCP_LOG_LEVEL": "ERROR"
      }
    }
  }
}

Using Docker

{
  "mcpServers": {
    "awslabs.security-agent-mcp-server": {
      "command": "docker",
      "args": [
        "run", "-i", "--rm",
        "-e", "AWS_REGION=us-east-1",
        "-e", "AWS_ACCESS_KEY_ID",
        "-e", "AWS_SECRET_ACCESS_KEY",
        "-e", "AWS_SESSION_TOKEN",
        "awslabs/security-agent-mcp-server:latest"
      ]
    }
  }
}

Environment Variables

Available Regions

See AWS documentation for available regions.

Available Tools

Setup

Code Review (orchestrated)

Remediation

Full API Access

Usage Flows

Code Review (source scan)

1. setup_check()              → verify readiness
2. setup()                    → provision resources (one-time)
3. start_security_scan(path=".")
4. get_scan_status()          → poll until COMPLETED
5. get_scan_findings()        → retrieve findings

Penetration Test

1. setup_check() → setup()   → one-time
2. call_api("CreateTargetDomain", {targetDomainName, verificationMethod})
3. call_api("VerifyTargetDomain", {targetDomainId})
4. call_api("CreatePentest", {agentSpaceId, title, assets: {endpoints: [...]}, serviceRole})
5. call_api("StartPentestJob", {agentSpaceId, pentestId})
6. Poll: call_api("BatchGetPentestJobs", {agentSpaceId, pentestJobIds})
7. call_api("ListFindings", {agentSpaceId, pentestJobId})

Any Operation

1. get_api_guide()            → see all operations + docs link
2. call_api(operation, params) → execute

Required IAM Permissions

These permissions are needed on your AWS credentials (the identity running the MCP server):

For setup (one-time)

• iam:CreateRole, iam:PutRolePolicy (if creating a new service role)
• s3:CreateBucket, s3:PutPublicAccessBlock, s3:PutLifecycleConfiguration (if creating a new bucket)
• sts:GetCallerIdentity
• securityagent:CreateAgentSpace, securityagent:UpdateAgentSpace
• securityagent:ListAgentSpaces, securityagent:BatchGetAgentSpaces

For code scanning

• s3:PutObject
• securityagent:CreateCodeReview, securityagent:StartCodeReviewJob
• securityagent:BatchGetCodeReviewJobs, securityagent:StopCodeReviewJob
• securityagent:ListFindings, securityagent:BatchGetFindings
• securityagent:StartCodeRemediation, securityagent:BatchDeleteCodeReviews

For pentesting and other operations

Add SecurityAgent permissions as needed for your use case. See How AWS Security Agent works with IAM for details on available actions.

Service Role

During setup, the server creates an IAM service role SecurityAgentScanRole (if one doesn't already exist). If an existing role is found on the agent space, it can be reused after validating its permissions.

The service role is assumed by the SecurityAgent service to read your uploaded code:

• Trust policy: securityagent.amazonaws.com service principal
• Permissions: S3 read on scan bucket, CloudWatch Logs write

Note: An S3 bucket is used to temporarily store source code for scanning. The MCP server sets a 30-day lifecycle policy on buckets it creates — uploaded content is automatically deleted. If you use your own bucket, consider adding a lifecycle rule to manage storage costs.

Contributing

Contributions are welcome! Please see the main repository's CONTRIBUTING.md for guidelines.

License

This project is licensed under the Apache License 2.0. See the LICENSE file for details.