AWS Agent Registry is Amazon’s centralized, governed catalog for discovering, approving, and reusing AI agents, tools, skills, and MCP servers, and it reached general availability on August 31, 2026, as part of Amazon Bedrock AgentCore. Google, Microsoft, and MuleSoft have each shipped a similar registry for their own platforms within the same year, and none of the four, including AWS’s, tells you whether the agent’s answer is still built on correct data. That second question is what Atlan, the context layer for AI, answers underneath whichever registry approved the agent in the first place.
The Atlan MCP server exposes governed business context, ownership, and lineage to any MCP-compatible agent, including one approved through AWS Agent Registry itself. A registry entry tells you an agent exists and cleared review. It says nothing about whether the metric it just quoted still means what it meant last quarter.
- A single catalog for agents, tools, skills, MCP servers, and custom resources across an organization
- A two-plane architecture that separates the full inventory from what’s actually approved for use
- An approval workflow with audit logging, not a check on the accuracy of what an agent knows
| What it is | Governed catalog for agents, tools, skills, and MCP servers |
| Key benefit | Cuts duplicate agent-building across teams lacking shared visibility |
| Best for | Enterprises running AgentCore agents across teams or business units |
| GA date | August 31, 2026 (preview since April 2026) |
| Regions | US East (N. Virginia), US West (Oregon), Europe (Ireland), Asia Pacific (Tokyo), Asia Pacific (Sydney) |
| Core components | Governance plane, discovery plane, approval workflow, hybrid search, auto-detection, audit log |
What is AWS Agent Registry?
Permalink to “What is AWS Agent Registry?”AWS Agent Registry is a managed AWS service, part of Amazon Bedrock AgentCore, that gives enterprises a single place to register, review, and discover the agents, tools, skills, and MCP servers running across their organization. It reached general availability on August 31, 2026, four months after entering public preview in April 2026, and the original bedrock-agentcore API namespace retires September 17, 2026, in favor of the agent-registry namespace introduced at GA.
The service runs on two complementary planes. AWS describes the governance plane as the “authoritative store for resources within its defined scope, regardless of their lifecycle state.” The discovery plane is “what consumers interact with day to day,” presenting only what cleared the approval bar, which matters for anyone building a semantic layer for AI agents: governance knows everything that exists, discovery is what a builder actually sees.
This is a narrower, more operational cousin of the compliance-focused AI registries built for regulatory inventory: a discovery and reuse problem, not an EU AI Act or NIST inventory problem, and conflating the two confuses how enterprises scope their AI agent governance programs. It’s also distinct from a model registry, which tracks ML model artifacts rather than agents. For the category, see what is an AI agent registry; for what’s being cataloged, see what is an AI agent and the types of AI agents enterprises register.
A registry entry passing approval is a claim about identity and process, not about whether the data underneath is current, the agent sprawl problem’s less-discussed sibling: knowing an agent exists doesn’t tell you whether what it knows is still true.
How does AWS Agent Registry work?
Permalink to “How does AWS Agent Registry work?”Four things determine how a resource becomes a discoverable, governed catalog entry: what can be registered, how it’s approved, how it’s found, and what gets logged.
What you can register
Permalink to “What you can register”AWS Agent Registry accepts four resource types. MCP servers register with their tools, resources, and prompts, relevant if you’re exposing enterprise data through MCP. A2A agent cards register agent-to-agent capabilities, per Atlan’s agent interoperability protocols rundown. Skills register as markdown definitions with a code package, per Atlan’s agent skills registry. Everything else registers as a custom resource via any valid JSON descriptor.
The approval lifecycle
Permalink to “The approval lifecycle”Every record moves through the same states: DRAFT, PENDING_APPROVAL, then APPROVED or REJECTED, with DEPRECATED as retirement. Admins configure the workflow via Amazon EventBridge, so compliance checks run before anything reaches the discovery plane, part of the broader set of AI agent risks and guardrails, a different question from whether the AI agent identity behind a record reasons from accurate context.
Search and discovery
Permalink to “Search and discovery”Once approved, the registry indexes a resource for hybrid search, semantic and lexical, across the catalog. MCP-compatible IDEs, including Kiro and Claude Code, connect natively. The MCP architecture deep dive covers what a compliant server implements.
| Object type | What it captures | Lifecycle states |
|---|---|---|
| MCP server | Tools, resources, and prompts via MCP | DRAFT → PENDING_APPROVAL → APPROVED / REJECTED → DEPRECATED |
| A2A agent | Agent-to-agent capability cards | Same four-state path |
| Skill | Markdown skill definition plus code package | Same four-state path |
| Custom resource | Any valid JSON descriptor | Same four-state path |
Approval confirms someone looked at a resource, not that it’s still correct, the gap a context repository closes.
What does AWS Agent Registry integrate with?
Permalink to “What does AWS Agent Registry integrate with?”A handful of integrations connect AWS Agent Registry to the rest of an AWS agent stack, each part of the broader AI control plane an enterprise assembles around its agents. AgentCore Gateway auto-syncs discovered endpoints. Amazon EventBridge emits a Registry Record State Change event on every state change, letting an organization wire approval logic or a human reviewer onto the registry. Amazon Quick surfaces approved agents, MCP servers, and skills on its Integrations page once connected. AWS IAM Identity Center handles authentication through OAuth or IAM-based auth, with AWS PrivateLink keeping the registry reachable without exposing it publicly.
Two more features matter for anyone worried about agents nobody signed off on. Auto-detection addresses shadow AI: enable endpoint detection once at the AWS Organization level, and the registry automatically detects agents and MCP servers running on AgentCore across every account. Every action, registration, approval, rejection, deprecation, gets a full audit trail in AWS CloudTrail, the kind of record that matters for teams managing GDPR compliance for AI agents. Custom metadata schemas, cost center, data classification, SLA tier, feed the same kind of agent context graph elsewhere in the stack.
Detection and audit logging answer “does this agent exist and who’s accountable.” They don’t answer “is the data this agent just used context-fresh,” which matters once an agent is making decisions, not just running.
Why did AWS build this now?
Permalink to “Why did AWS build this now?”According to Gartner (April 2026), the average global Fortune 500 enterprise will run more than 150,000 AI agents by 2028, up from fewer than 15 in 2025, and only 13% believe they have the right agent governance in place.
The problem shows up in team complaints before analyst reports: dozens of agents built by different teams within months, nobody left who can say what half of them do. That’s the failure mode a registry exists to prevent, not that agents get built too fast, but that nobody keeps a shared record of what exists.
Southwest Airlines is AWS’s headline reference for the GA launch. Lauren Woods, CIO/EVP at Southwest Airlines: “We went from dozens of agents and tools scattered across multiple technology teams with no shared record of what existed to a single, governed catalog that the entire organization trusts […] Our developers now find approved capabilities through semantic search in seconds instead of rebuilding what another team already built.”
Syngenta’s Sandeep Rayasa, Enterprise Architect for Data and AI: “We register, review, and approve every capability before sharing it across the organization, retaining clear ownership, versioning, and control over security and access.” Amdocs’ Ron Dublero, Chief Software Architect for Cognitive Core, called the registry “a trusted control plane for managing large-scale agent ecosystems.”
Every quote is about knowing what exists and trusting it was reviewed. Nothing in AWS’s launch materials claims the registry verifies that an approved agent’s knowledge is still accurate. That’s not an oversight. It’s outside the problem a catalog solves.
How does AWS Agent Registry compare to Google, Microsoft, and MuleSoft’s registries?
Permalink to “How does AWS Agent Registry compare to Google, Microsoft, and MuleSoft’s registries?”AWS is one of four major platforms to ship an agent registry within roughly a year, a pattern worth naming rather than a single vendor’s feature. Google’s Agent Registry, part of the Gemini Enterprise Agent Platform (the 2026 rebrand of Vertex AI), reached GA in late July 2026 as a centralized catalog in that platform’s “govern” layer. Microsoft’s answer is the unified registry inside Microsoft Agent 365, GA since May 1, syncing Copilot Studio agents automatically. MuleSoft’s Agent Fabric takes a more explicitly cross-cloud stance: its “Agent Scanners” auto-discover agents and MCP servers across Bedrock, Vertex AI, and other platforms.
| Platform | Registry | Status | Scope |
|---|---|---|---|
| AWS | AWS Agent Registry | GA, 2026-08-31 | Bedrock AgentCore-centric, extensible via custom JSON resources |
| Microsoft | Agent 365 registry | GA, 2026-05-01 | Microsoft 365 / Copilot Studio-centric |
| Agent Registry | GA, late July 2026 | Gemini Enterprise Agent Platform-centric | |
| MuleSoft | Agent Fabric Agent Registry | GA | Explicitly cross-cloud (Bedrock, Vertex AI, others) |
MuleSoft’s scanners and AWS’s own openness to custom JSON resources suggest the walls aren’t absolute. But scoped-by-default is the honest read: an agent approved in AWS Agent Registry has no native standing in Microsoft’s, and vice versa, unless someone builds the bridge by hand. Four platforms, four governance planes, and an enterprise running agents on more than one gets one inventory per cloud, each blind to what the others approved, why context portability matters more than picking the right one. See agent registry vs model registry for how a registry differs from an ML model registry.
What can’t AWS Agent Registry tell you?
Permalink to “What can’t AWS Agent Registry tell you?”An approved record certifies a narrow set of things: the resource exists, it passed schema validation against the MCP or A2A protocol, it cleared the configured approval workflow, and it has a known owner and version. That’s real, and it is not the same claim as certifying that the resource is correct.
Here’s the concrete version. A skill can clear every approval gate and still answer using a metric definition finance and product teams define differently, or one that changed last quarter without its owner finding out. Nothing in the schema, workflow, or CloudTrail log checks for that, a distinct discipline from context versioning for AI agents. The registry validated identity. It never validated knowledge.
This isn’t a gap unique to AWS. Google’s, Microsoft’s, and MuleSoft’s registries share the same structural limit, because none is built to answer a question about data freshness or business meaning. Trust in what ran and trust in what it knew are two different audits, and a registry, any registry, only ever performs the first.
How Atlan approaches context for agents registered in AWS Agent Registry
Permalink to “How Atlan approaches context for agents registered in AWS Agent Registry”An agent that’s identity-approved in AWS Agent Registry can still be authorized to run against context nobody has separately verified. That’s the gap the enterprise context layer closes: it supplies what the registry was never designed to check.
The Atlan MCP server exposes governed enterprise context, ownership, certification, lineage, and the glossary terms and metrics defined in the Active Ontology, to any MCP-compatible agent, the same governed access described in how to give AI agents access to enterprise data. The pattern: deploy the MCP server, register it with AgentCore Gateway as a tool source, publish the resource in AWS Agent Registry, and agents on AgentCore find it through the registry and call it through Gateway. What comes back is access-controlled, current business context resolved at the moment the agent asks, the practical version of making AI agents context-aware.
The registry’s approval gate and this context resolution answer different questions at the same call. The registry confirms the agent is allowed to run and the tool is who it claims to be. The response confirms it reflects what’s true in the business right now, which determines whether the output is trustworthy once someone acts on it. Enterprise-ready AI agents need both audits running, and it’s why context engineering is a production discipline that runs continuously.
Real stories from real customers: MCP servers in production
Permalink to “Real stories from real customers: MCP servers in production”"Atlan captures Workday's shared language to be leveraged by AI via its MCP server. As part of Atlan's AI labs, we're co-building the semantic layer that AI needs."
Joe DosSantos, VP Enterprise Data & Analytics, Workday
"Atlan is much more than a catalog of catalogs. It's more of a context operating system. Atlan enabled us to easily activate metadata for everything from discovery in the marketplace to AI governance to data quality to an MCP server delivering context to AI models."
Sridher Arumugham, Chief Data & Analytics Officer, DigiKey
Why an approved agent still needs a context layer
Permalink to “Why an approved agent still needs a context layer”AWS Agent Registry, and its Google, Microsoft, and MuleSoft equivalents, solved the half of agent governance that’s about inventory: what exists, who owns it, whether it was reviewed. Gartner expects that problem to get two orders of magnitude worse by 2028. The half none of the four touch is whether what’s inside an approved agent is still true, a question this cohort addresses directly in does an agent registry need a context layer. An agent can pass every gate and still be wrong, why governing agents across multiple clouds means more than reconciling four catalogs. It means building an enterprise context layer underneath all of them, one that holds steady no matter which registry approved the agent that day.
FAQs about AWS Agent Registry
Permalink to “FAQs about AWS Agent Registry”1. What is AWS Agent Registry?
Permalink to “1. What is AWS Agent Registry?”AWS Agent Registry is a managed AWS service, part of Amazon Bedrock AgentCore, that provides a centralized, governed catalog for discovering, approving, and reusing AI agents, tools, skills, and MCP servers across an organization. It reached general availability on August 31, 2026.
2. How does AWS Agent Registry prevent agent sprawl?
Permalink to “2. How does AWS Agent Registry prevent agent sprawl?”It gives every team a searchable catalog of what already exists, so builders reuse an approved agent instead of rebuilding one another team created. Auto-detection also surfaces unregistered “shadow” agents running on AgentCore infrastructure.
3. Is AWS Agent Registry a runtime security tool?
Permalink to “3. Is AWS Agent Registry a runtime security tool?”No. It’s a governance and discovery service, not a runtime security product. Runtime enforcement, authentication, and access control are handled separately by AgentCore Gateway and AWS IAM Identity Center.
4. What’s the difference between AWS Agent Registry and an agent hub or marketplace?
Permalink to “4. What’s the difference between AWS Agent Registry and an agent hub or marketplace?”A registry is a system of record for internal governance: it tracks ownership, approval status, and version for resources an organization controls. A hub or marketplace is built for consumption and distribution across boundaries, without the same approval-workflow and audit-trail requirements.
5. Is AWS Agent Registry generally available, or still in preview?
Permalink to “5. Is AWS Agent Registry generally available, or still in preview?”It’s generally available as of August 31, 2026. It launched in preview in April 2026 under the bedrock-agentcore namespace, retired September 17, 2026, in favor of the agent-registry namespace introduced at GA.
6. Which AWS regions support AWS Agent Registry?
Permalink to “6. Which AWS regions support AWS Agent Registry?”At GA, it’s available in US East (N. Virginia), US West (Oregon), Europe (Ireland), Asia Pacific (Tokyo), and Asia Pacific (Sydney).
7. Does AWS Agent Registry verify that an agent’s data or answers are accurate?
Permalink to “7. Does AWS Agent Registry verify that an agent’s data or answers are accurate?”No. It verifies that a resource exists, passed schema validation, cleared approval, and has a known owner and version. It doesn’t check whether the business data or definitions an agent reasons over are still correct, a separate problem a context layer addresses.
Sources
Permalink to “Sources”- Manage agents, tools, and skills at scale with AWS Agent Registry, AWS Machine Learning Blog, 2026. https://aws.amazon.com/blogs/machine-learning/manage-agents-tools-and-skills-at-scale-with-aws-agent-registry/
- AWS Agent Registry: Discover and manage agents, tools, and resources, Amazon Bedrock AgentCore Developer Guide, 2026. https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/registry.html
- Key capabilities, Amazon Bedrock AgentCore Developer Guide, 2026. https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/registry-key-capabilities.html
- AWS Agent Registry for centralized agent discovery and governance is now available in Preview, AWS, April 2026. https://aws.amazon.com/about-aws/whats-new/2026/04/aws-agent-registry-in-agentcore-preview
- Release notes for Amazon Bedrock AgentCore, AWS, 2026. https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/release-notes.html
- Gartner Identifies Six Steps to Manage AI Agent Sprawl, Gartner Newsroom, April 2026. https://www.gartner.com/en/newsroom/press-releases/2026-04-28-gartner-identifies-six-steps-to-manage-artificial-intelligence-agent-sprawl
- AWS targets AI agent sprawl with new Bedrock Agent Registry, InfoWorld, 2026. https://www.infoworld.com/article/4157183/aws-targets-ai-agent-sprawl-with-new-bedrock-agent-registry.html
- AWS Launches Agent Registry in Preview to Govern AI Agent Sprawl across Enterprises, InfoQ, April 2026. https://www.infoq.com/news/2026/04/aws-agent-registry-preview/
- Microsoft Agent 365, now generally available, expands capabilities and integrations, Microsoft Security Blog, May 2026. https://www.microsoft.com/en-us/security/blog/2026/05/01/microsoft-agent-365-now-generally-available-expands-capabilities-and-integrations/
- MuleSoft launches Agent Fabric to tackle agent sprawl and unify enterprise AI workflows, CIO, 2026. https://www.cio.com/article/4063090/mulesoft-launches-agent-fabric-to-tackle-agent-sprawl-and-unify-enterprise-ai-workflows.html
- Google AI Agents Get Seven-Day Runtime and Memory Bank, Enterprise DNA, 2026. https://enterprisedna.co/resources/news/google-gemini-agent-platform-memory-runtime-identity-ga-2026/
