Usage instructions (AMI with CloudFormation delivery)

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 the region (us-east-1).
2. Fill in the parameters:

1. Acknowledge that the stack creates IAM resources and create the stack. It takes 10 to 15 minutes: the instance boots, generates its TLS certificate, creates the mcp_agent Postgres role, publishes its CA certificate, and a Lambda-backed custom resource registers the MCP target on the gateway and waits for it to become READY.

Stack outputs

First call: tools/list

The gateway uses IAM authorization (SigV4). Calls are signed with the IAM role the caller already runs under: the instance profile of the EC2 instance or the execution role of the Lambda function hosting your agent, or your own signed-in role when testing from a workstation. Grant that role bedrock-agentcore:InvokeGateway on the gateway ARN. Nothing is typed into the product.

The example below runs anywhere an IAM role is present (EC2 instance profile, Lambda execution role, ECS task role, or a signed-in AWS CLI session). The AWS SDK picks up that role on its own:

import json, boto3
from botocore.auth import SigV4Auth
from botocore.awsrequest import AWSRequest
import urllib.request

gateway_url = boto3.client("cloudformation").describe_stacks(StackName="<stack>")["Stacks"][0]["Outputs"]
gateway_url = next(o["OutputValue"] for o in gateway_url if o["OutputKey"] == "GatewayUrl")

body = json.dumps({"jsonrpc": "2.0", "id": 1, "method": "tools/list"})
req = AWSRequest(method="POST", url=gateway_url, data=body, headers={"Content-Type": "application/json"})
SigV4Auth(boto3.Session().get_credentials(), "bedrock-agentcore", "us-east-1").add_auth(req)
print(urllib.request.urlopen(urllib.request.Request(gateway_url, data=body.encode(), headers=dict(req.headers), method="POST")).read().decode())

From an agent framework, point the MCP client at GatewayUrl with SigV4 signing for service bedrock-agentcore; every major SDK signs with the same IAM role lookup. The AIDB tools (semantic_kb_search, SQL and schema tools) appear in the tool list. If you prefer JWT, add a Cognito or OIDC authorizer to the gateway after creation; the template creates it with AuthorizerType: AWS_IAM.

What is inside

• Database: EPAS 18 on Ubuntu 24.04, database edb, extensions aidb, edb_vectorplus, edb_cloneschema and edb_job_scheduler (for dbms_job and dbms_scheduler); SPL and database links are built in. Local admin: sudo -iu enterprisedb psql -d edb over SSM.
• Tools: EDB*Plus (edbplus), JDBC, ODBC and OCL connectors, Language Pack (plperl, plpython3, pltcl), LDAP Sync. Database links to Oracle (edb_dblink_oci) need an Oracle Instant Client, which you install yourself.
• MCP database role: mcp_agent is created at first boot from the Secrets Manager secret the stack generates. The gateway presents it to the MCP server; nothing is baked into the AMI.
• Semantic Knowledge Base: a Semantic KB named semkb over schema public is created at build time with live auto processing. Tables you add to public are indexed as rows arrive; semantic_kb_search works from the first table.
• Embeddings model: nomic-embed-text-v1.5 (GGUF, llama.cpp) is pre-downloaded in the image. No internet access is needed at runtime.
• Access: shell via AWS Systems Manager Session Manager only. There is no SSH key and password login is disabled. Port 22 listens for AWS Marketplace scanning but the security group has no inbound rule for it.

Security group recommendations

None required. The template creates two security groups: the gateway connection group (no inbound) and the instance group, which accepts TCP 8765 only from the gateway connection group.

Data and updates

The Postgres data directory is on a separate EBS volume, encrypted with the AWS managed key. The volume is kept when the stack is deleted, and it keeps costing money until you delete it.

• Updating the stack (for example to a new AMI version) replaces the instance. The old instance is stopped, which shuts Postgres down cleanly, and the new instance mounts the same volume. Your databases, models, agents and Semantic KB carry over.
• Changing the subnet to one in another Availability Zone cannot move the volume. Do not do this on a stack that holds data.
• To start a new stack from the data of a deleted one, enter the old volume's ID as ExistingDataVolumeId.
• Postgres configuration files outside the data directory (/etc/edb-as) come from the AMI and are not kept.

Remove

Delete the stack. The custom resource removes the gateway target and its AgentCore Identity entry; the gateway, instance, secrets and IAM roles are deleted with the stack. The data volume stays. Delete it in the EC2 console (Volumes) when you no longer need the data.

Version 1.2. Listing: AWS Marketplace. Support: techsupport.enterprisedb.com.