acrun

acrun

A lightweight deployment tool for AWS Bedrock AgentCore Runtime.

acrun is inspired by lambroll (Lambda) and ecspresso (ECS). It focuses on deployment operations and works with infrastructure managed elsewhere (Terraform/CloudFormation).

Features

• Simple workflow: init, diff, deploy, invoke
• Jsonnet/JSON config with useful native functions
• Infrastructure separation; no scaffolding, just deploy
• CI/CD friendly with clear exit codes and colored diffs

Install

• Go: go install github.com/mashiike/acrun/cmd/acrun@latest
• Binaries: download from GitHub Releases (GoReleaser; darwin/linux/windows, amd64/arm64)

Quick Start

1. Initialize from an existing runtime

acrun init --agent-runtime-name my-agent --format jsonnet
# writes ./agent_runtime.jsonnet (or json via --format json)

1. Edit agent_runtime.jsonnet

{
  agentRuntimeName: "my-agent",
  roleArn: "arn:aws:iam::123456789012:role/MyAgentRole",
  agentRuntimeArtifact: {
    containerConfiguration: {
      containerUri: std.native('ecrImageUri')('my-agent', std.native('env')('IMAGE_TAG', 'latest')),
    },
  },
  environmentVariables: { stage: std.native('env')('STAGE', 'dev') },
}

1. See the diff against remote

acrun diff --exit-code            # exit code 2 when different
acrun diff --ignore '.agentRuntimeVersion'   # jq query to ignore fields

1. Deploy to an endpoint (non-DEFAULT)

acrun deploy --endpoint-name staging
# creates/updates runtime and the endpoint, then points it to the new version

1. Invoke for a quick check

echo '{"inputText":"ping"}' | acrun invoke --endpoint-name staging

Commands

• init: Fetch runtime by name and write agent_runtime.jsonnet or agent_runtime.json.
  • Flags: --agent-runtime-name, --qualifier <endpoint|version>, --format json|jsonnet, --force-overwrite
• diff: Compare local file with remote runtime (version or endpoint).
  • Flags: --qualifier <endpoint|version> (default: current), --ignore <jq>, --exit-code
• deploy: Create/update runtime and update or create the specified endpoint.
  • Flags: --endpoint-name <name> (required; cannot be DEFAULT), --dry-run
• invoke: Call the deployed agent runtime with a payload.
  • Flags: --payload, --content-type, --accept, --endpoint-name (default: current), plus MCP/trace headers (--mcp-proto-version, --mcp-session-id, --runtime-session-id, --runtime-user-id, --baggage, --trace-id, --trace-parent, --trace-state)
• render: Print normalized config from local file.
  • Flags: --format json|jsonnet
• delete: Delete the runtime (safe by default).
  • Flags: --force, --dry-run
• rollback: Point an endpoint to an older version.
  • Flags: --endpoint-name <name> (cannot be DEFAULT), --version <n> (default: current-1), --dry-run

Global flags:

• --agent-runtime <path>: Path to config file (defaults: agent_runtime.jsonnet or agent_runtime.json in CWD)
• --tfstate <url|path>: Terraform state location; same as ACRUN_TFSTATE
• --log-level <debug|info|warn|error> and --log-format <text|json>

Exit codes:

• diff returns exit code 2 (via --exit-code) when there are differences; other commands use 0/1.

Configuration

acrun reads agent_runtime.jsonnet or agent_runtime.json in the working directory by default. Fields are lowerCamelCase to align with AWS API. You can use Jsonnet to compose per-environment configs.

Jsonnet native functions are provided for convenience. See the detailed section below for usage and examples.

Example: Terraform integration

local tf = std.native('tfstate');
{
  roleArn: tf('aws_iam_role.agent_runtime.arn'),
  agentRuntimeArtifact: {
    containerConfiguration: {
      containerUri: tf('aws_ecr_repository.agent.repository_url') + ':' + std.native('env')('IMAGE_TAG','latest'),
    },
  },
  networkConfiguration: {
    networkMode: 'VPC',
    vpcConfig: {
      subnetIds: [ tf('aws_subnet.private["az-a"].id'), tf('aws_subnet.private["az-b"].id') ],
      securityGroupIds: [ tf('aws_security_group.agent_runtime.id') ],
    },
  },
}

Terraform state locations supported: local files, S3 (s3://...), HTTP/HTTPS, GCS (gs://...), Azure Blob (azurerm://...).

Jsonnet Native Functions

acrun provides several native functions for use in agent_runtime.jsonnet, following Jsonnet's camelCase naming convention:

callerIdentity()

Returns AWS caller identity information from STS GetCallerIdentity API.

• Returns: Object with account, arn, and userId fields (camelCase).

Example:

local identity = std.native('callerIdentity')();
local accountId = identity.account;

{
  roleArn: 'arn:aws:iam::' + accountId + ':role/MyRole',
  agentRuntimeArtifact: {
    containerConfiguration: {
      containerUri: accountId + '.dkr.ecr.us-west-2.amazonaws.com/my-agent:latest',
    },
  },
  environmentVariables: {
    awsAccountId: accountId,
    awsArn: identity.arn,
    awsUserId: identity.userId,
  },
}

env(name, default)

Returns environment variable value, or default if not set.

Example:

{
  environmentVariables: {
    stage: std.native('env')('STAGE', 'dev'),
  },
}

mustEnv(name)

Returns environment variable value, or raises an error if not set.

Example:

{
  environmentVariables: {
    apiKey: std.native('mustEnv')('API_KEY'),
  },
}

ecrImageUri(repositoryName, imageTag)

Resolves ECR container image URI and verifies the repository exists.

• Parameters:
  • repositoryName: ECR repository name (e.g., "acrun/sample-mcp")
  • imageTag: Image tag (e.g., "latest", "v1.0.0")
• Returns: Full ECR image URI string (e.g., "123456789012.dkr.ecr.us-west-2.amazonaws.com/acrun/sample-mcp:latest").
• Features:
  • Verifies repository exists in ECR using DescribeRepositories
  • Automatically obtains registry ID and repository URI
  • Constructs the complete ECR URI with the specified tag

Example:

{
  agentRuntimeArtifact: {
    containerConfiguration: {
      // Automatically resolves account/region and verifies image exists
      containerUri: std.native('ecrImageUri')('my-agent', 'v1.0.0'),
    },
  },
}

Use with environment variables:

local tag = std.native('env')('IMAGE_TAG', 'latest');
{
  agentRuntimeArtifact: {
    containerConfiguration: {
      containerUri: std.native('ecrImageUri')('my-agent', tag),
    },
  },
}

tfstate(address)

Looks up values from Terraform state file.

• Prerequisites: Set --tfstate flag or ACRUN_TFSTATE environment variable to point to your Terraform state file (local path or S3 URL).
• Parameters:
  • address: Terraform resource address (e.g., "aws_iam_role.agent_runtime.arn", "data.aws_subnet.private.id")
• Returns: The value from the Terraform state at the specified address.

Example:

# Set tfstate location
export ACRUN_TFSTATE="s3://my-bucket/terraform.tfstate"
# or
acrun deploy --tfstate s3://my-bucket/terraform.tfstate

// In agent_runtime.jsonnet
local tfstate = std.native('tfstate');

{
  // Reference IAM role managed by Terraform
  roleArn: tfstate('aws_iam_role.agent_runtime.arn'),

  agentRuntimeArtifact: {
    containerConfiguration: {
      // Reference ECR repository
      containerUri: tfstate('aws_ecr_repository.agent.repository_url') + ':latest',
    },
  },

  networkConfiguration: {
    networkMode: 'VPC',
    vpcConfig: {
      // Reference VPC subnets
      subnetIds: [
        tfstate('aws_subnet.private["az-a"].id'),
        tfstate('aws_subnet.private["az-b"].id'),
      ],
      // Reference security groups
      securityGroupIds: [
        tfstate('aws_security_group.agent_runtime.id'),
      ],
    },
  },
}

Supported state locations:

• Local file: /path/to/terraform.tfstate
• S3: s3://bucket/path/to/terraform.tfstate
• HTTP/HTTPS: https://example.com/terraform.tfstate
• Google Cloud Storage: gs://bucket/path/to/terraform.tfstate
• Azure Blob Storage: azurerm://container/path/to/terraform.tfstate

Endpoint Semantics

• current qualifier resolves the version backing the named endpoint and is used by default in diff/invoke.
• DEFAULT endpoint is reserved; deploy/rollback against DEFAULT are intentionally blocked. Use your own endpoint names (e.g., dev, staging, prod).

Examples

See _examples/agent/ for a minimal agent project and example configs.

Build from source

go build ./cmd/acrun

Acknowledgements

acrun's design is inspired by:

• lambroll: https://github.com/fujiwara/lambroll
• ecspresso: https://github.com/kayac/ecspresso

License

See LICENSE.