Skip to content

Limit on Bedrock AgentCore Gateway Targets?

0

Hi,

We appear to be hitting some kind of limit related to the number of targets configured on an AgentCore gateway.

Our initial approach was to take a large OpenAPI specification, convert it to MCP, and integrate it with Amazon Quick. That failed ('Creation failed'). To troubleshoot, we took a smaller subset of the same OpenAPI spec, added only those endpoints as targets, and that integration worked successfully.

Based on that result, we split the original large OpenAPI spec into multiple smaller subsets, resulting in roughly 50 targets in total. All of these targets were added successfully to the AgentCore gateway.

The problem occurs when we try to integrate this gateway with Amazon Quick. The creation step fails with the message "Creation failed", but we cannot find any corresponding logs or diagnostics that explain the failure. To clarify, all of the targets are configured under a single gateway, and we are still using only one gateway MCP URL for the integration.

Is anyone aware of a limit on the number of targets, or any other constraints that might cause this behavior?

Any guidance would be appreciated.

2 Answers
1

Yes, there are specific limits related to Amazon Bedrock AgentCore Gateway targets that may be relevant to your situation.

The default quota is 100 targets per gateway, which is adjustable through the Service Quotas console. Since you mentioned having roughly 50 targets, you're within this limit, so the target count itself shouldn't be the direct cause of your integration failure.

However, there are additional related quotas that could be contributing factors:

  • 1,000 tools per target (adjustable)
  • Maximum inline schema size of 1 MB (adjustable)
  • Maximum S3 payload schema size of 10 MB (adjustable)
  • Tool name character limit of 256 characters (adjustable)
  • Gateway invocation timeout of 15 minutes (adjustable)

Given that your integration works with a smaller subset but fails with the full set of 50 targets, the issue might be related to the cumulative size or complexity of all the schemas and tools across those targets, rather than just the target count. The total schema size across all targets, the number of tools defined within those targets, or the overall complexity of the configuration could be exceeding limits during the integration process with Amazon Q.

I'd recommend checking the total size of your schemas across all targets and the total number of tools being exposed. You may need to request quota increases through the Service Quotas console, or further optimize your API specification to reduce the overall footprint of your gateway configuration.
Sources
Quotas for Amazon Bedrock AgentCore - Amazon Bedrock AgentCore

answered 2 months ago
EXPERT
reviewed 2 months ago
0

To add to what the agent said -- you're likely hitting the 1MB inline schema limit unless you've already adjusted that service quota. If your mentioned 1.2MB is the raw OpenAPI spec, you're probably looking at a fair bit more than that as the MCP tool definition generated from them typically is larger (mostly because each tool gets its own full JSON schema with expanded parameters, no references).

You could consider using S3-based schemas instead of inline schemas for your target (as that allows for 10MB of payload), or splitting into 2-3 gateways, or requesting a service quota increase.

AWS
answered 2 months ago

You are not logged in. Log in to post an answer.

A good answer clearly answers the question and provides constructive feedback and encourages professional growth in the question asker.