convez.ai
MCP ServerIntegração & APIsAvançado
tmanmidwest

Agente de Suporte Saviynt com AWS Bedrock

tmanmidwest/AgentAWSBedwithMCP

Agente inteligente Bedrock que descobre ferramentas Saviynt em tempo real

Ver no GitHub
InicialPluginScore · de 0 a 100

MCP Server que implementa um agente de atendimento ao cliente executado no AWS Bedrock AgentCore, integrando dinamicamente ferramentas do servidor Saviynt via protocolo MCP. Descobre e utiliza automaticamente novos tools conforme disponibilizados, sem necessidade de reconfiguração.

Instalação

$npx convez-ai tmanmidwest/AgentAWSBedwithMCP

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
28%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Atendimento automatizado de suporte ao cliente com acesso a dados Saviynt
  • Automação de consultas Q&A integradas ao Saviynt Gateway
  • Orquestração de fluxos de trabalho com descoberta dinâmica de ferramentas

Veredito

Solução robusta e bem arquitetada para integrar Claude/Bedrock com Saviynt em produção. Setup automatizado com setup.py é prático, mas exige conhecimento sólido de AWS, Docker e MCP para troubleshooting.

README

A tradução para o português deste README ainda está na fila.

Saviynt Support Agent — AWS Bedrock AgentCore

A customer support / Q&A agent that dynamically discovers and uses tools from the Saviynt MCP server, deployed on Amazon Bedrock AgentCore Runtime.


Architecture

User / Client
     │
     ▼  HTTPS
AWS Bedrock AgentCore Runtime  (ARM64 container, port 8080)
     │  FastAPI  /invocations
     ▼
Strands Agent  (Claude Sonnet via Amazon Bedrock)
     │  MCP  Streamable-HTTP
     ▼
Saviynt MCP Server  (tool discovery + invocation)

The agent dynamically discovers every tool the Saviynt MCP server advertises at runtime — no hardcoded tool list needed. When Saviynt adds new MCP tools, the agent picks them up automatically.


Quickstart — automated setup

The fastest way to get running is the interactive wizard. It collects your credentials, creates all AWS resources, builds the Docker image, and deploys the agent in one go:

python setup.py

The only things you need before running it:

RequirementNotes
AWS CLI v2 configuredaws configure — needs ECR, IAM, Bedrock, AgentCore permissions
Docker Desktop runningWith buildx available
Python 3.xNo extra packages needed to run setup.py
Saviynt MCP URL + credentialsFrom your Saviynt admin

The wizard will ask for everything else interactively and save a .env file for local development.


Manual prerequisites (if not using setup.py)

RequirementNotes
AWS accountWith AgentCore, Bedrock, and ECR permissions
IAM roleBedrockAgentCoreExecutionRole (see below)
Saviynt MCP URLe.g. https://<tenant>.saviyntcloud.com/mcp
Saviynt credentialsOAuth client-credentials or API key
Docker + buildxARM64 cross-compilation or native ARM64 host
AWS CLI v2Configured with the right profile / region
Python 3.12For local dev / testing only

Required IAM permissions for BedrockAgentCoreExecutionRole

Attach these managed policies (or equivalent inline):

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Action": [
        "bedrock:InvokeModel",
        "bedrock:InvokeModelWithResponseStream"
      ],
      "Resource": "arn:aws:bedrock:*::foundation-model/anthropic.*"
    },
    {
      "Effect": "Allow",
      "Action": [
        "logs:CreateLogGroup",
        "logs:CreateLogStream",
        "logs:PutLogEvents"
      ],
      "Resource": "arn:aws:logs:*:*:*"
    },
    {
      "Effect": "Allow",
      "Action": [
        "xray:PutTraceSegments",
        "xray:PutTelemetryRecords"
      ],
      "Resource": "*"
    }
  ]
}

The trust policy must allow bedrock-agentcore.amazonaws.com:

{
  "Version": "2012-10-17",
  "Statement": [{
    "Effect": "Allow",
    "Principal": { "Service": "bedrock-agentcore.amazonaws.com" },
    "Action": "sts:AssumeRole"
  }]
}

Project structure

saviynt-support-agent/
├── agent.py              # FastAPI + Strands agent (main entrypoint)
├── requirements.txt      # Python dependencies
├── Dockerfile            # ARM64 container image
├── deploy.sh             # Build → ECR → AgentCore deployment script
├── agentcore/
│   └── agentcore.json    # AgentCore Runtime metadata
└── tests/
    └── test_local.py     # Local smoke-test (no AWS deployment needed)

Environment variables

Set these before running locally or via AWS Secrets Manager / Parameter Store in production.

VariableRequiredDescription
SAVIYNT_MCP_URL✅Full URL of the Saviynt MCP endpoint
SAVIYNT_CLIENT_ID⚠️OAuth client ID (or API key)
SAVIYNT_CLIENT_SECRET⚠️OAuth client secret
SAVIYNT_TOKEN_URL⚠️OAuth token endpoint for 2LO flow
AWS_DEFAULT_REGION✅e.g. us-east-1
BEDROCK_MODEL_IDoptionalDefault: us.anthropic.claude-sonnet-4-5

Auth modes — the agent supports two Saviynt MCP auth modes:

  1. OAuth 2.0 Client Credentials (2LO): set all three SAVIYNT_* vars.
  2. Static Bearer / API key: set only SAVIYNT_CLIENT_ID.

Local development

# 1. Install dependencies
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

# 2. Export environment variables
export SAVIYNT_MCP_URL="https://your-tenant.saviyntcloud.com/mcp"
export SAVIYNT_CLIENT_ID="your-client-id"
export SAVIYNT_CLIENT_SECRET="your-secret"
export SAVIYNT_TOKEN_URL="https://your-tenant.saviyntcloud.com/oauth/token"
export AWS_DEFAULT_REGION="us-east-1"

# 3. Run the server
python agent.py

# 4. In another terminal, run the smoke tests
python tests/test_local.py

Deploy to AWS Bedrock AgentCore

# Edit EXECUTION_ROLE_ARN in deploy.sh, then:
chmod +x deploy.sh
./deploy.sh

The script will:

  1. Build an ARM64 Docker image using docker buildx
  2. Push it to Amazon ECR
  3. Create (or update) the AgentCore Runtime agent

Invoke from anywhere

aws bedrock-agentcore invoke-agent-runtime \
  --region us-east-1 \
  --agent-runtime-arn arn:aws:bedrock-agentcore:us-east-1:123456789:agent-runtime/saviynt-support-agent \
  --payload '{"input":{"prompt":"What pending approvals do I have?","session_id":"abc123"}}'

Extending the agent

Add conversation memory — integrate AgentCore Memory for persistent session context across invocations. See the Strands Agents docs for AgentCoreMemory.

Add a Gateway — if you want to expose additional Lambda or API Gateway tools alongside the Saviynt MCP server, create an AgentCore Gateway and add it as a second MCP target.

Change the model — set BEDROCK_MODEL_ID to any model available in your region, e.g. us.anthropic.claude-opus-4-5.

Update the system prompt — edit SYSTEM_PROMPT in agent.py to tune the agent's persona, tone, or escalation rules for your Saviynt use-case.

Ver o original no GitHub

Bom para

Relacionados