Blocks.ai publishes this comparison and sells in this market. Every claim about another product links to a public source.
Amazon Bedrock AgentCore vs DIY: a managed AWS platform or an in-house build
Amazon Bedrock AgentCore
AWS platform for building, deploying, and operating AI agents
Build it yourself (DIY)
Building agent connections and controls in-house from open protocols, existing infrastructure, and open-source tools
Short answer
Amazon Bedrock AgentCore is a product: AWS’s managed platform for building, deploying, and operating agents, made of modular services you can use together or independently.Source 1 DIY is not a product: your teams connect and govern agents themselves, on open protocols such as MCP and A2A that leave authorization logic to the implementer.Source 2, Source 3
Where each one sits
Six layers of running AI agents at a company, and what each product’s own public sources say it covers.
These aren’t the same kind of product
Amazon Bedrock AgentCore
DIY
At a glance
What each one is
Amazon Bedrock AgentCore
Amazon Bedrock AgentCore is AWS’s platform for building, deploying, and operating agents with any framework and foundation model.Source 1 Its modular services, usable together or independently, include a serverless Runtime for hosting agents, Gateway, Identity, Policy, Observability, and AWS Agent Registry.Source 1, Source 6, Source 11, Source 27, Source 28
Build it yourself (DIY)
DIY means connecting and governing agents without buying an agent platform: open protocols wired together in-house, existing infrastructure stretched to cover agents, an in-house platform, self-hosted open-source frameworks, or no central approach; Uber, for one, built its own MCP registry and proxy gateway.Source 17
The differences that matter
Agent registry
Amazon Bedrock AgentCoreAWS Agent Registry can list agents, MCP servers, tools, and skills behind an approval workflow, and has been generally available since August 2026.Source 11, Source 19
DIYA2A’s specification prescribes no standard API for curated registries, and its docs say one requires deploying and maintaining a registry service.Source 16
The official MCP Registry, in preview, doesn’t support private servers; Uber and Pinterest run their own.Source 17, Source 21, Source 29 With AWS Organizations, Agent Registry can currently auto-detect only AgentCore Runtimes and Gateways.Source 30
Identity and policy
Amazon Bedrock AgentCoreAgent identities are workload identities, and Policy can check each request through a Gateway before tool access, using rules in natural language or Cedar.Source 10, Source 27
DIYA2A leaves authorization logic to each server and MCP makes it optional; Uber’s gateway applies policies from its internal Access Control System.Source 2, Source 17, Source 24
MCP’s roadmap says MCP authorization assumes a person with a browser at consent time.Source 31
Sharing across accounts and companies
Amazon Bedrock AgentCoreA registry can be shared with other AWS accounts through AWS RAM; accounts outside your AWS Organization must accept an invitation.Source 13
DIYA2A is designed for agents built by different companies on separate servers; each server authorizes requests under its own policies.Source 2, Source 32
AWS says its record-level condition keys don’t apply to registry search and list, which return summaries of all matching records.Source 13
For security teams
What a security review asks, answered from each vendor’s public documentation.
Full comparison
18 criteria in five groups. Every cell links to its source, or says no public source answers it.
| Amazon Bedrock AgentCore | DIY | |
|---|---|---|
| What it is | ||
| What it is and who it’s for | AWS’s platform for building, deploying, and operating agents with any framework and model, for moving agents from proof of concept to production.Source 1, Source 6 | An in-house build on open protocols: A2A, an open standard for communication between agent systems, and MCP, which A2A calls complementary.Source 2 |
| Maturity | Generally available since October 2025.Source 19 Policy GA in March 2026; AWS Agent Registry GA in August 2026.Source 19 | Varies by part: MCP’s latest specification revision is 2026-07-28.Source 3 The official MCP Registry is in preview.Source 21 |
| Control | ||
| Agent registry and discovery | AWS Agent Registry can list agents, MCP servers, tools, and skills behind an approval workflow, including ones on premises or in other clouds.Source 1, Source 11 | Build your own: A2A sets no standard API for curated registries.Source 16 The MCP Registry, in preview, recommends your own for private servers.Source 21 |
| Identity and access control | Agent identities are workload identities.Source 10 Hosted agents take IAM SigV4 by default, or JWTs from any OAuth 2.0 provider.Source 22, Source 23 Policy can check Gateway traffic.Source 27 | In A2A, identity is established at the HTTP layer, and authorization logic is implementation-specific.Source 2, Source 18 MCP authorization is optional.Source 24 |
| Ownership, policy, and revocation | Gateway policies, in natural language or Cedar, set which tools an agent may call and when.Source 27 One control to disable an agent everywhere: not publicly documented. | MCP leaves consent and authorization flows to implementers.Source 3 Uber starts MCP servers and tools disabled; Pinterest gives non-experimental servers an owning team.Source 17, Source 29 |
| Audit log and observability | Metrics, spans, and logs go to CloudWatch, OpenTelemetry-compatible.Source 28 CloudTrail can log Gateway API calls and logs Registry control-plane calls.Source 30, Source 36 | A2A docs advise auditing significant events and tracing, for example with OpenTelemetry.Source 18 Pinterest’s MCP servers log inputs, outputs, and exceptions.Source 29 |
| Connection | ||
| How agents connect | Agents can run in serverless Runtime, or elsewhere behind a Gateway that can forward to their endpoint URL.Source 1, Source 12 Outbound-only connections: not publicly documented. | MCP servers on Streamable HTTP expose an HTTP endpoint; A2A agents declare a URL.Source 2, Source 34 A standard outbound-only binding: not publicly documented. |
| Agents across organizations | A registry can be shared with other AWS accounts via AWS RAM to discover, publish, or administer.Source 13 Runtime agents can be opened to principals in other accounts.Source 35 | A2A is designed for agents from different companies on separate servers.Source 32 Its docs highly recommend API management across organizations.Source 18 |
| Protocol support | Runtime agents can serve HTTP, MCP, A2A, or AG-UI.Source 5 Gateway supports MCP 2025-03-26, 2025-06-18, 2025-11-25, and 2026-07-28 for MCP targets.Source 39 | MCP defines stdio and Streamable HTTP transports and allows custom ones.Source 40 A2A maps to JSON-RPC, gRPC, and HTTP/REST bindings.Source 2 |
| Frameworks, models, and clouds supported | Runtime works with CrewAI, LangGraph, LlamaIndex, Google ADK, OpenAI Agents SDK, Strands Agents, and custom frameworks.Source 1 | A2A gives agents built on different frameworks, languages, or vendors a common language.Source 2 Google’s ADK says it is deployment-agnostic.Source 14 |
| Operations | ||
| Deployment options and data residency | AWS says cross-region inference can move Memory, Policy, and Evaluations prompts out of the primary Region; AgentCore may store content to improve your service.Source 1, Source 41 | Wherever you run it: Pinterest optimized for MCP servers in its internal cloud.Source 29 A2A leaves protecting stored data to your policies.Source 18 |
| Compliance attestations | AWS lists AgentCore as FedRAMP (Class C and Class D) compliant.Source 37, Source 38 In scope for SOC 1, 2, and 3 reports; ISO and CSA STAR compliant; HIPAA eligible.Source 19, Source 37 | Sits with the implementer: A2A docs name GDPR, CCPA, and HIPAA; MCP leaves access controls and data protections to implementers.Source 3, Source 18 |
| Support and SLA | AWS says the Amazon Bedrock SLA applies to AgentCore.Source 6 Basic Support is included for all AWS customers.Source 42 | Depends on the component: MCP SDKs are tiered partly by maintenance commitments; the official MCP Registry, in preview, gives no uptime guarantees.Source 43, Source 44 |
| Time and effort to get running | AWS’s quickstart scaffolds, tests, and deploys one agent with the AgentCore CLI.Source 4 It needs an AWS account and Node.js 20 or later.Source 4 | A curated A2A registry is a service you deploy and maintain.Source 16 Pinterest built a unified deployment pipeline after new MCP servers took too much setup.Source 29 |
| Pricing model and public prices | Consumption-based per service, no minimum fee.Source 25 Examples: Policy $0.000025 per authorization request; Runtime v1 CPU $0.0895 per vCPU-hour.Source 25 | A2A and MCP are openly licensed specifications, A2A under Apache 2.0.Source 2, Source 26 Build and running costs are not publicly documented. |
| Building | ||
| Agent building tools | Write the agent loop in Python with a framework such as Strands, LangGraph, or Google ADK and deploy it to Runtime, or use the managed Harness.Source 1, Source 4 | Open-source frameworks such as LangGraph and Google’s ADK build and deploy agents.Source 8, Source 14 The A2A project hosts SDKs in six languages.Source 45 |
| Model access | Model-agnostic, AWS says: models in or outside Amazon Bedrock, including OpenAI, Gemini, Claude, Nova, Llama, and Mistral.Source 6 | Your choice per agent: Google’s ADK, for example, is optimized for Gemini and model-agnostic.Source 14 |
| Integrations and ecosystem | Gateway has 1-click integrations such as Salesforce, Slack, Jira, Asana, and Zendesk, and can import AWS Partner tools bought through AWS Marketplace.Source 6, Source 46 | The official MCP Registry, in preview, offers a REST API to discover MCP servers and an OpenAPI spec other registries can implement.Source 21 |
Which to choose
Choose Amazon Bedrock AgentCore if
- You want hosting, identity, policy, and a registry as managed AWS services you can adopt together or one at a time.Source 1
- You want a registry with an approval workflow, so builders discover only records that meet your organization’s criteria.Source 11, Source 20
- Your agents already span AWS accounts, and you want one registry shared through AWS RAM, with record-level controls on reads and writes.Source 13
- You need vendor attestations: AWS lists AgentCore as HIPAA eligible, in scope for SOC reports, and ISO and CSA STAR compliant.Source 19, Source 37
Choose DIY if
- You want your own routing and security logic on agent traffic, as Pinterest applies to MCP servers in its internal cloud.Source 29
- You want access checks from systems you already run, as Uber and Pinterest do with internal access control and mesh identities.Source 17, Source 29
- You want your own review rules, such as Uber’s: every MCP server and tool starts disabled until reviewed and enabled.Source 17
- You want no agent platform contract: A2A and MCP are openly licensed specifications you can build on.Source 2, Source 26
Questions buyers ask
Can Amazon Bedrock AgentCore manage agents that run outside AWS?
AWS says its Registry works with agents and MCP servers on premises or in other clouds, and that AgentCore Identity manages identities for self-hosted agents.Source 1, Source 9 Automatic discovery through AWS Organizations can currently create records only for AgentCore Runtimes and Gateways.Source 30
Does Amazon Bedrock AgentCore support MCP and A2A?
Can Amazon Bedrock AgentCore be self-hosted?
A self-hosted or customer-operated edition is not publicly documented. AWS presents AgentCore as a managed service, offered in AWS Regions including AWS GovCloud (US-West).Source 1, Source 6, Source 19 A DIY build runs where you deploy it; A2A’s docs leave protecting its data to your own security policies.Source 18
How do the costs compare?
How we compare
Read the full methodEvery claim on this page links to a public source. Where none answers a question, the page says so.
We re-check every fact at least every 90 days. This page was last checked .
Something wrong or out of date? Tell us and we’ll correct it.
Sources
53 public sources, each with the date we checked it. Every one opens in a new tab.
Source 1: Overview - Amazon Bedrock AgentCore (Developer Guide) Back:abcdefghijklmnopqr
Source 2: Agent2Agent (A2A) Protocol Specification Back:abcdefghijklmnopq
Source 3: Specification - Model Context Protocol 2026-07-28 Back:abcd
Source 4: Get started with Amazon Bedrock AgentCore Back:abcde
Source 5: Understand the AgentCore Runtime service contract Back:abc
Source 10: Provide identity and credential management for agent applications with Amazon Bedrock AgentCore Identity Back:abcde
Source 11: AWS Agent Registry: Discover and manage agents, tools, and resources Back:abcdef
Source 12: HTTP passthrough targets - AgentCore Gateway Back:abcd
Source 13: Sharing a registry across accounts with AWS RAM Back:abcdef
Source 15: Integrating with Model Context Protocol (MCP) - Keycloak Back:ab
Source 17: Designing MCP Gateway Uber's MCP Management Platform - Uber Blog Back:abcdefgh
Source 18: Enterprise Features - A2A Protocol Back:abcdefghij
Source 19: Release notes - Amazon Bedrock AgentCore Back:abcdefg
Source 20: Concepts and terminology - AWS Agent Registry Back:ab
Source 21: The MCP Registry - Model Context Protocol Back:abcde
Source 22: Authenticate and authorize with Inbound Auth and Outbound Auth Back:abc
Source 24: Authorization - Model Context Protocol specification 2026-07-28 Back:abc
Source 26: modelcontextprotocol/modelcontextprotocol LICENSE (GitHub) Back:abcd
Source 27: Policy in Amazon Bedrock AgentCore: Control Agent Interactions Back:abcde
Source 28: Observe your agent applications on Amazon Bedrock AgentCore Observability Back:abc
Source 29: Building an MCP Ecosystem at Pinterest - Pinterest Engineering Blog Back:abcdefg
Source 33: Use interface VPC endpoints (AWS PrivateLink) with Amazon Bedrock AgentCore Back to text
Source 34: Streamable HTTP - Model Context Protocol specification 2026-07-28 Back:ab
Source 35: Resource-based policies for Amazon Bedrock AgentCore Back:ab
Source 36: Log Amazon Bedrock AgentCore Gateway API calls with CloudTrail Back:ab
Source 37: Compliance validation for Amazon Bedrock AgentCore Back:abcd
Source 38: Federal Risk and Authorization Management Program (FedRAMP) - Services in Scope - Amazon Web Services Back:ab
Source 40: Transports - Model Context Protocol specification 2026-07-28 Back to text
Source 41: Cross-region inference in AgentCore Memory, Policy in AgentCore, and AgentCore Evaluations Back to text
Source 44: MCP Registry Aggregators - Model Context Protocol Back to text
Source 46: Amazon Bedrock AgentCore Gateway: A secure AI gateway for agents, tools, and models Back to text
Source 47: Add observability to your Amazon Bedrock AgentCore resources Back to text
Source 48: Deploy A2A servers in AgentCore Runtime Back to text