OpenADA MCP AgentCore - Accessibility Tools for Bedrock

## How It Works

The product runs as a managed AgentCore container and connects to either the hosted OpenADA service or a private OpenADA ECS deployment. Customers provide the OpenADA MCP URL and, when needed, a protected-deployment API key. No OpenADA database or crawler needs to be installed inside the AgentCore runtime.

The architecture is simple and stateless:

• Amazon Bedrock AgentCore Runtime invokes the container via MCP over stateless Streamable HTTP
• The container forwards requests to the configured OpenADA MCP endpoint
• OpenADA performs accessibility checks, language analysis, scans, and reporting

## Key Capabilities for AI Agents

With OpenADA MCP AgentCore, your AI agents can:

• **Check public pages for accessibility findings** - Identify WCAG issues using axe-core powered analysis
• **Detect language-quality issues** - LanguageTool-compatible analysis for content clarity
• **Start bounded same-host website scans** - Asynchronous crawls with configurable scope
• **Monitor scan progress and retrieve final reports** - Durable tracking of scan state
• **Browse public site scores and scan history** - Access the OpenADA public directory
• **Compare page findings over time** - Track accessibility improvement across deployments

## Deployment Options

OpenADA MCP AgentCore supports two backend configurations:

• **Hosted OpenADA service** - Connect to the public OpenADA endpoint at openada.us for immediate evaluation of public pages without infrastructure setup
• **Private OpenADA ECS deployment** - Point to your own OpenADA infrastructure for organization-controlled data paths, API policies, and network isolation

## AWS Integration

• **Amazon Bedrock AgentCore Runtime** - Managed container execution with built-in scaling
• **IAM and SigV4 authentication** - AWS-native security at the runtime boundary
• **VPC networking support** - Deploy in private subnets with security group controls for internal OpenADA endpoints
• **ARM64 container architecture** - Runs on the AgentCore MCP container contract
• **No custom SigV4 implementation required** - Authentication is handled by the AgentCore boundary

## Configuration

The container requires minimal configuration through environment variables:

• **OPENADA_MCP_URL** (required) - The OpenADA MCP endpoint URL
• **OPENADA_API_KEY** (optional) - Bearer key for protected OpenADA deployments

The gateway listens on port 8000 with a health check at GET /ping and MCP requests at POST /mcp.

## Security and Privacy

• The gateway forwards requests only to the configured OpenADA MCP URL
• API keys are supplied as runtime environment secrets, never stored in source control
• AWS SigV4 credentials remain with the invoking IAM identity
• The public OpenADA service accepts public URLs only with bounded crawl protections
• Private deployments can enforce API keys, CORS restrictions, and allowed scan hosts

## Important Limitations

OpenADA combines axe-core accessibility checks, LanguageTool-compatible language analysis, public scan history, and developer-friendly APIs. Results are engineering guidance for remediation and are not a legal determination or certification of ADA compliance. Automated accessibility findings do not cover every WCAG or ADA requirement. Site crawls are asynchronous and bounded by the OpenADA deployment's scan settings.

## Who This Is For

Public agencies, developers, and accessibility teams who want to integrate accessibility checking into AI agent workflows without operating a separate MCP server or building a custom integration.