AWS launches Agent Registry for runtime discovery of MCP tools and agents
In short: AWS introduced Agent Registry, a managed catalog service for Model Context Protocol (MCP) servers, agents, and agent skills. Instead of hardcoding every tool into an orchestrator, agents can now search an approved, versioned catalog at runtime and build a client from the returned descriptor. A reference sample demonstrates a Discover → Instantiate → Execute loop using Amazon Bedrock AgentCore Gateway and AgentCore runtime, with Anthropic Claude Sonnet as the foundation model.
This summary was generated automatically by AI from AWS Machine Learning's publication. It is our own text, not a copy of the original — facts, figures and quotes belong to the source, linked above and below.
What changed?
- 1New AWS Agent Registry service for cataloging MCP servers, agents, and agent skills
- 2Search returns record identity, descriptors, and protocol metadata, but never invocation credentials
- 3Records go through a DRAFT → PENDING_APPROVAL → APPROVED workflow with IAM-authorized actions
- 4Orchestrator builds an ephemeral inner agent from discovered tools per request, then tears it down
- 5Hybrid keyword/semantic search via search_discoverable_registry_records
- 6Reference sample uses Amazon Bedrock AgentCore Gateway, AgentCore runtime, Cognito, Lambda, Secrets Manager, and Anthropic Claude Sonnet via Bedrock
Why it matters
Hardcoded tool lists in agent orchestrators become unmanageable as enterprise API surfaces grow past a few dozen tools, requiring code changes and redeploys for every new integration; runtime discovery decouples tool registration from orchestrator deployment and lets new capabilities go live after registry approval alone.
What it means for AI agents and contact centers
For teams building voice agents or call-center automation that call many backend systems (order lookup, CRM, pricing, support), this pattern shows how to avoid hardwiring every API into the agent and instead let it search a governed catalog at call time, potentially easing integration maintenance as the number of connected systems grows, though it is AWS/Bedrock-specific infrastructure.
🧪 Worth Testing
Worth evaluating if agent architecture already uses AWS Bedrock/AgentCore or MCP, to see whether runtime tool discovery reduces engineering overhead versus maintaining a fixed tool list per agent.
Sources
- AWS Machine LearningOfficialPrimary sourceOriginal article →„Discover tools at runtime with AWS Agent Registry“20 Oct 2026, 19:22
- Published by source
- 20 Oct 2026, 19:22
- Found by our system
- 20 Oct 2026, 19:47
- Summary generated
- 20 Oct 2026, 19:50
This article was written by AI from the original source. Facts, numbers and prices come from the source; missing values are marked “Not specified”. Legal notice, copyright and privacy