Add stateless_http configuration option

BEDROCK_AGENTCORE_DEPLOYMENT.md

@@ -0,0 +1,59 @@
+# Purple AI MCP - Deployment on Amazon Bedrock Agent Core
+
+This guide will take you through deploying the Purple AI MCP Server to Amazon Bedrock AgentCore. 
+
+## Prerequisites
+
+**Obtain a Sentinelone Singularity Operations Center console token**
+
+This can be found in Policy & Settings → User Management → Service Users in your console. Currently, this server only supports tokens that have access to a single Account or Site. If you need to access multiple sites, you will need to run multiple MCP servers with Account-specific or Site-specific tokens.
+
+**Prepare Environment Variables** 
+
+```bash
+PURPLEMCP_CONSOLE_BASE_URL=https://your-console.sentinelone.net
+PURPLEMCP_CONSOLE_TOKEN=your-token
+MCP_MODE=streamable-http 
+PURPLEMCP_STATELESS_HTTP=True
+```
+
+**Prepare AWS Environment**
+
+### VPC Configuration
+
+ **Outbound internet access** - Outbound internet access should be permitted through an Internet Gateway or NAT Gateway. \
+ **Security Groups** - Allow outbound HTTPS on port 443 to connect to the Sentinelone service. 
+
+It is important to note that Purple AI MCP does not include built-in authentication. For network exposed AWS deployments, ensure the MCP Server is placed behind an Application Load Balancer (ALB) with the appropriate timeout settings. Detailed information on production setups can be found [here](PRODUCTION_SETUP.md#cloud-load-balancer-setup).
+
+### IAM Configuration
+
+When deploying Purple AI MCP via AWS Marketplace a 'default' service role will be automatically created. To use a customer-managed service role reference the IAM Policy below.
+
+[Trust Policy](bedrock-agentcore-trust-policy.json)
+[IAM Policy](bedrock-agentcore-iam-policy.json)  
+
+## Agent Core Configuration Settings
+
+When configuring the MCP in Bedrock Agent core, set the following.
+
+1. **Name** - Give the deployment a suitable name and description.
+2. **Agent Source** - Default (ECR) as this is automatically populated via AWS Marketplace. 
+3. **IAM Permissions** - Default service role or create a custom service role based on the IAM Configuration step [above](#iam-configuration).
+4. **Protocol** - Set to MCP.
+5. **Inbound Identity Type** - Select JWT (JSON Web Tokens) or IAM.
+    - **JWT** - Requires IdP bearer tokens to invoke the agent.
+    - **IAM** - Requires relevant IAM Permissions to invoke the agent.
+6. **Security Type** - Select VPC and set the relevant VPC, Subnets & Security Groups.
+7. **Advanced Settings** - Add environment variables as noted in the [prerequisites](#prerequisites).
+8. Click **Host Agent** to complete the setup.
+
+The Agent will take around ~1-2 minutes to become active.
+
+## Useful Links
+
+[Production VPC Setup for Purple AI MCP](PRODUCTION_SETUP.md) \
+[AWS Bedrock AgentCore Overview](https://aws.amazon.com/bedrock/agentcore/) \
+[Granting a VPC Internet Access via Internet Gateway](https://docs.aws.amazon.com/vpc/latest/userguide/VPC_Internet_Gateway.html) \
+[Creating custom IAM Roles](https://docs.aws.amazon.com/IAM/latest/UserGuide/id_roles_create_for-service.html) 
+

README.md

@@ -58,6 +58,19 @@ docker run -p 8000:8000 \
   purple-mcp:latest
 ```
 
+### Using Amazon Bedrock AgentCore
+```bash
+# Subscribe to Purple AI MCP Server via AWS Marketplace
+
+#Prepare Environment Variables
+PURPLEMCP_CONSOLE_BASE_URL=https://your-console.sentinelone.net
+PURPLEMCP_CONSOLE_TOKEN=your-token
+MCP_MODE=streamable-http 
+PURPLEMCP_STATELESS_HTTP=True
+```
+Follow instructions for Amazon Bedrock AgentCore Deployment [here](BEDROCK_AGENTCORE_DEPLOYMENT.md)
+
+
 For production deployments, see [Deployment Guide](DOCKER.md).
 
 **Note:** Purple AI MCP does not include built-in authentication. For network-exposed deployments, place it behind a reverse proxy or load balancer. See [Production Setup](PRODUCTION_SETUP.md) for cloud load balancer configurations (AWS ALB, GCP Cloud Load Balancing, Azure Application Gateway) or nginx examples for self-hosted deployments.
@@ -189,6 +202,7 @@ We suggest you **do not** expose Purple AI MCP on a network at this time, as the
 ## Environment Variables
 - `PURPLEMCP_CONSOLE_TOKEN` - Service user token (Account or Site level)
 - `PURPLEMCP_CONSOLE_BASE_URL` - Console URL (e.g., https://console.sentinelone.net)
+- `PURPLEMCP_STATELESS_HTTP` - For use with deployment in Amazon Bedrock Agent Core - Detailed instructions can be found [here](BEDROCK_AGENTCORE_DEPLOYMENT.md)
 
 
 ## Development

bedrock-agentcore-iam-policy.json

@@ -0,0 +1,100 @@
+{
+    "Version": "2012-10-17",
+    "Statement": [
+        {
+            "Sid": "ECRImageAccess",
+            "Effect": "Allow",
+            "Action": [
+                "ecr:BatchGetImage",
+                "ecr:GetDownloadUrlForLayer"
+            ],
+            "Resource": [
+                "arn:aws:ecr:us-east-1:709825985650:repository/sentinelone/purple-ai-mcp-server"
+            ]
+        },
+        {
+            "Effect": "Allow",
+            "Action": [
+                "logs:DescribeLogStreams",
+                "logs:CreateLogGroup"
+            ],
+            "Resource": [
+                "arn:aws:logs:{{region}}:{{accountId}}:log-group:/aws/bedrock-agentcore/runtimes/*"
+            ]
+        },
+        {
+            "Effect": "Allow",
+            "Action": [
+                "logs:DescribeLogGroups"
+            ],
+            "Resource": [
+                "arn:aws:logs:{{region}}:{{accountId}}:log-group:*"
+            ]
+        },
+        {
+            "Effect": "Allow",
+            "Action": [
+                "logs:CreateLogStream",
+                "logs:PutLogEvents"
+            ],
+            "Resource": [
+                "arn:aws:logs:{{region}}:{{accountId}}:log-group:/aws/bedrock-agentcore/runtimes/*:log-stream:*"
+            ]
+        },
+        {
+            "Sid": "ECRTokenAccess",
+            "Effect": "Allow",
+            "Action": [
+                "ecr:GetAuthorizationToken"
+            ],
+            "Resource": "*"
+        },
+        {
+            "Effect": "Allow",
+            "Action": [
+                "xray:PutTraceSegments",
+                "xray:PutTelemetryRecords",
+                "xray:GetSamplingRules",
+                "xray:GetSamplingTargets"
+            ],
+            "Resource": [
+                "*"
+            ]
+        },
+        {
+            "Effect": "Allow",
+            "Resource": "*",
+            "Action": "cloudwatch:PutMetricData",
+            "Condition": {
+                "StringEquals": {
+                    "cloudwatch:namespace": "bedrock-agentcore"
+                }
+            }
+        },
+        {
+            "Sid": "GetAgentAccessToken",
+            "Effect": "Allow",
+            "Action": [
+                "bedrock-agentcore:GetWorkloadAccessToken",
+                "bedrock-agentcore:GetWorkloadAccessTokenForJWT",
+                "bedrock-agentcore:GetWorkloadAccessTokenForUserId"
+            ],
+            "Resource": [
+                "arn:aws:bedrock-agentcore:{{region}}:{{accountId}}:workload-identity-directory/default",
+                "arn:aws:bedrock-agentcore:{{region}}:{{accountId}}:workload-identity-directory/default/workload-identity/hosted_agent_s26ce-*"
+            ]
+        },
+        {
+            "Sid": "BedrockModelInvocation",
+            "Effect": "Allow",
+            "Action": [
+                "bedrock:InvokeModel",
+                "bedrock:InvokeModelWithResponseStream"
+            ],
+            "Resource": [
+                "arn:aws:bedrock:*::foundation-model/*",
+                "arn:aws:bedrock:{{region}}:{{accountId}}:*"
+            ]
+        }
+    ]
+}

bedrock-agentcore-trust-policy.json

@@ -0,0 +1,21 @@
+{
+    "Version": "2012-10-17",
+    "Statement": [
+        {
+            "Sid": "AssumeRolePolicy",
+            "Effect": "Allow",
+            "Principal": {
+                "Service": "bedrock-agentcore.amazonaws.com"
+            },
+            "Action": "sts:AssumeRole",
+            "Condition": {
+                "StringEquals": {
+                    "aws:SourceAccount": "{{accountId}}"
+                },
+                "ArnLike": {
+                    "aws:SourceArn": "arn:aws:bedrock-agentcore:{{region}}:{{accountId}}:*"
+                }
+            }
+        }
+    ]
+}

docker-compose.yml

@@ -37,6 +37,7 @@ services:
       MCP_MODE: streamable-http
       MCP_HOST: 0.0.0.0
       MCP_PORT: 8000
+      PURPLEMCP_STATELESS_HTTP: ${PURPLEMCP_STATELESS_HTTP}
       # Required: SentinelOne Console configuration
       PURPLEMCP_CONSOLE_BASE_URL: ${PURPLEMCP_CONSOLE_BASE_URL}
       PURPLEMCP_CONSOLE_TOKEN: ${PURPLEMCP_CONSOLE_TOKEN}

src/purple_mcp/cli.py

@@ -71,24 +71,30 @@ def _setup_logging(verbose: bool) -> None:
 
 
 def _apply_environment_overrides(
+    transport_mode: str | None,
     sdl_api_token: str | None,
     graphql_service_token: str | None,
     console_base_url: str | None,
     graphql_endpoint: str | None,
     alerts_graphql_endpoint: str | None,
+    stateless_http: bool | None,
 ) -> None:
     """Apply CLI argument values to environment variables.
 
     Populates os.environ with provided CLI arguments, allowing command-line
     configuration to override environment variable defaults.
 
     Args:
+        transport_mode: MCP transport mode to use.
         sdl_api_token: SDL API authentication token
         graphql_service_token: GraphQL service authentication token
         console_base_url: Base URL for the console
         graphql_endpoint: GraphQL endpoint path
         alerts_graphql_endpoint: Alerts GraphQL endpoint path
+        stateless_http: Uses true stateless mode (new transport per request)
     """
+    if transport_mode:
+        os.environ[f"{ENV_PREFIX}TRANSPORT_MODE"] = transport_mode
     if sdl_api_token:
         os.environ[f"{ENV_PREFIX}SDL_READ_LOGS_TOKEN"] = sdl_api_token
     if graphql_service_token:
@@ -102,6 +108,8 @@ def _apply_environment_overrides(
         and alerts_graphql_endpoint != "/web/api/v2.1/unifiedalerts/graphql"
     ):
         os.environ[f"{ENV_PREFIX}ALERTS_GRAPHQL_ENDPOINT"] = alerts_graphql_endpoint
+    if stateless_http:
+        os.environ[f"{ENV_PREFIX}STATELESS_HTTP"] = str(stateless_http)
 
 
 def _check_unsupported_config() -> None:
@@ -240,6 +248,7 @@ def _run_uvicorn(
     port: int,
     verbose: bool,
     allow_remote_access: bool,
+    stateless_http: bool,
 ) -> None:  # pragma: no cover - uvicorn is mocked in unit-tests
     """Run the HTTP/SSE transport using *uvicorn*."""
     # Validate host binding for security
@@ -258,7 +267,8 @@ def _run_uvicorn(
         from purple_mcp.server import app
 
         # Create the HTTP app and instrument it if Logfire is enabled
-        http_app = app.http_app(transport=transport)
+        # n.b. stateless_http has no effect if running in "sse" transport mode.
+        http_app = app.http_app(transport=transport, stateless_http=stateless_http)
         instrument_starlette_app(http_app)
 
         uvicorn.run(
@@ -280,21 +290,28 @@ def _run_mode(
     verbose: bool,
     no_banner: bool = False,
     allow_remote_access: bool = False,
+    stateless_http: bool = False,
 ) -> None:
     """Dispatch to the appropriate server runner for *mode*."""
     mode_normalised = mode.lower()
 
     runners: Mapping[str, Callable[[], None]] = {
         "stdio": lambda: _run_stdio(verbose, no_banner),
         "sse": lambda: _run_uvicorn(
-            "sse", host=host, port=port, verbose=verbose, allow_remote_access=allow_remote_access
+            "sse",
+            host=host,
+            port=port,
+            verbose=verbose,
+            allow_remote_access=allow_remote_access,
+            stateless_http=stateless_http,
         ),
         "streamable-http": lambda: _run_uvicorn(
             "streamable-http",
             host=host,
             port=port,
             verbose=verbose,
             allow_remote_access=allow_remote_access,
+            stateless_http=stateless_http,
         ),
     }
 
@@ -308,6 +325,7 @@ def _run_mode(
     "-m",
     type=click.Choice(VALID_MODES, case_sensitive=False),
     default="stdio",
+    envvar=f"{ENV_PREFIX}TRANSPORT_MODE",
     help="MCP transport mode to use",
 )
 @click.option(
@@ -365,6 +383,15 @@ def _run_mode(
     is_flag=True,
     help="Allow binding to non-loopback interfaces (SECURITY RISK: exposes unauthenticated tools)",
 )
+@click.option(
+    "--stateless-http",
+    is_flag=True,
+    envvar=f"{ENV_PREFIX}STATELESS_HTTP",
+    help=(
+        "Uses true stateless mode (new transport per request). This flag only has an effect if running in"
+        " http or streamable_http modes."
+    ),
+)
 def main(
     mode: str,
     host: str,
@@ -377,6 +404,7 @@ def main(
     verbose: bool,
     banner: bool,
     allow_remote_access: bool,
+    stateless_http: bool,
 ) -> None:
     """Purple MCP Server - AI monitoring and analysis tool."""
     _setup_logging(verbose)
@@ -385,11 +413,13 @@ def main(
     _check_unsupported_config()
 
     _apply_environment_overrides(
+        transport_mode=mode,
         sdl_api_token=sdl_api_token,
         graphql_service_token=graphql_service_token,
         console_base_url=console_base_url,
         graphql_endpoint=graphql_endpoint,
         alerts_graphql_endpoint=alerts_graphql_endpoint,
+        stateless_http=stateless_http,
     )
 
     settings = _create_settings()
@@ -406,6 +436,7 @@ def main(
         verbose=verbose,
         no_banner=not banner,
         allow_remote_access=allow_remote_access,
+        stateless_http=stateless_http,
     )
 
 

src/purple_mcp/config.py

@@ -9,7 +9,7 @@
 import logging
 import uuid
 from functools import lru_cache
-from typing import ClassVar, Final
+from typing import ClassVar, Final, Literal
 from urllib.parse import urlparse
 
 from pydantic import Field, field_validator
@@ -45,6 +45,8 @@
 PURPLE_AI_CONSOLE_VERSION_ENV: Final[str] = f"{ENV_PREFIX}PURPLE_AI_CONSOLE_VERSION"
 ENVIRONMENT_ENV: Final[str] = f"{ENV_PREFIX}ENV"
 LOGFIRE_TOKEN_ENV: Final[str] = f"{ENV_PREFIX}LOGFIRE_TOKEN"
+STATELESS_HTTP_ENV = f"{ENV_PREFIX}STATELESS_HTTP"
+TRANSPORT_MODE_ENV = f"{ENV_PREFIX}TRANSPORT_MODE"
 
 
 class Settings(BaseSettings):
@@ -170,6 +172,18 @@ class Settings(BaseSettings):
         validation_alias=LOGFIRE_TOKEN_ENV,
     )
 
+    stateless_http: bool | None = Field(
+        default=False,
+        description="Stateless mode (new transport per request)",
+        validation_alias=STATELESS_HTTP_ENV,
+    )
+
+    transport_mode: Literal["http", "streamable-http", "sse"] = Field(
+        default="sse",
+        description="Stateless mode (new transport per request)",
+        validation_alias=TRANSPORT_MODE_ENV,
+    )
+
     @field_validator("sentinelone_console_base_url")
     @classmethod
     def validate_console_base_url(cls, v: str) -> str: