AWS Agent Registry guide demonstrates dynamic discovery and invocation of tools and agents
AWS described an orchestrator written in Python that uses AWS Agent Registry to discover tools and agents based on a request and creates the corresponding clients. It obtains credentials separately.
AWS published an implementation guide for discovering tools and agents while processing a request. The example connects an orchestrator written in Python, an MCP server for order management and two agents using the A2A protocol. The orchestrator sends a natural-language query to AWS Agent Registry and uses the returned descriptions to create clients for the capabilities it finds.
Search results contain connection metadata, but neither credentials nor the authentication method. For the MCP server accessible through Amazon Bedrock AgentCore Gateway, the example obtains an OAuth token separately from Amazon Cognito. It calls agents using the A2A protocol through a boto3 client, which signs requests with the AWS credentials of the caller. For each request, a temporary internal agent is created with only the tools found; once processing is complete, the agent and clients are disposed of.
The example covers deployment, populating the registry, testing search, three demonstration scenarios and resource cleanup. With manual approval, a record moves through the DRAFT, PENDING_APPROVAL and APPROVED states; search returns only approved records. IAM permissions allow publishing to be separated from approval. AWS recommends testing search quality with representative queries and does not assume that a larger catalog will automatically improve result ranking.
Why it matters
The guide gives developers a concrete procedure for assembling a set of available capabilities for each request without hard-coding every tool into the orchestrator. It also shows that finding the right tool does not in itself resolve authorization to use it: the application must obtain credentials through a separate, trusted channel.
Two audiences, two different impacts
What this means
For individuals
Using the example, a developer can implement the creation of MCP and A2A clients based on the retrieved description and refresh the set of tools for each request.
For a business
For companies using tools from multiple teams, the design described offers a way to separate publishing and approving records through IAM permissions. This ensures that only approved records appear in search results.
DevelopmentCheck the original
Event sources
only one source so far · 1 publisher, 0 independent. We count feeds from the same owner only once.