EDB Postgres AI Agent Factory

EDB Postgres AI Agent Factory is EDB Postgres Advanced Server 18 with the AIDB extension, Vector Plus and the AIDB MCP server, delivered as one CloudFormation stack that also creates an Amazon Bedrock AgentCore Gateway and registers the MCP server on it. Agents call the gateway; the gateway reaches the instance over a private VPC Lattice connection. No port is opened to the internet and no SSH key is used.

LAUNCH

1. Choose Launch CloudFormation template on the product page and pick us-east-1.

2. Parameters. AmiId: filled in by AWS Marketplace, leave as is. VpcId: a VPC with outbound access (NAT gateway, transit gateway or VPC endpoints) to Secrets Manager, Systems Manager and Bedrock AgentCore. SubnetId: a subnet in that VPC; the instance and the gateway's private connection both use it. InstanceType: m6i.large, m6i.xlarge (default) or m6i.2xlarge; 4 vCPU and 16 GiB or more is recommended because the embeddings model runs on the CPU. GatewayName: name of the AgentCore Gateway, unique in the account and region. DataVolumeSize: GiB for the Postgres data volume (default 50). ExistingDataVolumeId: optional, the DataVolumeId output of a deleted stack to start from its data. SendTracesToCloudWatch (default true) and TelemetryRetentionDays (default 90): AIDB traces, logs and metrics to X-Ray and CloudWatch.

3. Acknowledge that the stack creates IAM resources and create it. It takes 10 to 15 minutes while the instance boots, creates the mcp_agent role and publishes its certificate, and a custom resource registers the MCP target on the gateway.

STACK OUTPUTS

GatewayUrl is the MCP endpoint your agent connects to. GatewayId is the gateway identifier. DataVolumeId is the EBS volume holding the Postgres data; it is kept when the stack is deleted. InstanceId is for a shell with AWS Systems Manager Session Manager (aws ssm start-session --target InstanceId). McpEndpoint is the private MCP URL on the instance, reachable only from the gateway. CredentialsSecretArn and CaCertificateSecretArn are the stack's Secrets Manager secrets.

FIRST CALL

The gateway uses IAM authorization (SigV4). Calls are signed with the IAM role the caller already runs under: the EC2 instance profile or Lambda execution role hosting your agent, or your own signed-in role from a workstation. Grant that role bedrock-agentcore:InvokeGateway on the gateway ARN. Nothing is typed into the product. From an agent framework, point the MCP client at GatewayUrl with SigV4 signing for service bedrock-agentcore. The AIDB tools (semantic_kb_search, SQL and schema tools) appear in tools/list. For JWT, add a Cognito or OIDC authorizer to the gateway after creation; the template creates it with AuthorizerType AWS_IAM. A worked Python example is in the full instructions linked below.

WHAT IS INSIDE

EPAS 18 on Ubuntu 24.04, database edb, extensions aidb, edb_vectorplus, edb_cloneschema and edb_job_scheduler; SPL and database links are built in. EDB*Plus, the JDBC, ODBC and OCL connectors, Language Pack and LDAP Sync are installed. A Semantic Knowledge Base named semkb over schema public is created at build time; tables you add to public are indexed as rows arrive. The embeddings model is in the image, so no internet access is needed at runtime. Shell access is through Session Manager only; there is no SSH key and password login is disabled. The template creates two security groups and opens nothing to the internet: the instance accepts TCP 8765 only from the gateway's connection.

DATA AND UPDATES

Postgres data lives on its own encrypted EBS volume that survives instance replacement and stack deletion. To upgrade, update the stack with the new version; the data volume moves to the new instance. To delete, delete the stack; the data volume stays until you delete it in the EC2 console.

Full instructions with the Python example and parameter table: https://edb-marketplace-templates-352127991267-us-east-1.s3.us-east-1.amazonaws.com/edb-postgres-ai/v1.2/usage.html