kagent with AWS Bedrock models: ModelConfig returns 404 model_not_found; is Bedrock supported?
I'm using kagent (kagent-dev/kagent) and want to run Claude Sonnet 3.7, DeepSeek, or Amazon Nova through AWS Bedrock. I created a ModelConfig CR with provider: OpenAI and baseUrl https://bedrock-runtime.us-west-2.amazonaws.com/openai/v1 (opens external site), but requests with anthropic.claude-3-7-sonnet-20250219-v1:0 fail with
Error code: 404 - model_not_found. Does kagent support Bedrock models directly, or is there a working workaround?Originally asked on GitHub: https://github.com/kagent-dev/kagent/issues/183
4 answers
As of the last update in the thread (December 2025), kagent does not have first-class support for Bedrock models/APIs; a contributor confirmed this and said contributions would be welcome. The issue was closed as completed after v0.1.11 added support for Anthropic models, but that did not resolve the Bedrock setup described below.
The configuration that produced the 404 error was:
yamlapiVersion: kagent.dev/v1alpha2 kind: ModelConfig metadata: name: bedrock-config namespace: default spec: apiKeySecret: kagent-bedrock apiKeySecretKey: AWS_API_KEY model: anthropic.claude-3-7-sonnet-20250219-v1:0 openAI: baseUrl: https://bedrock-runtime.us-west-2.amazonaws.com/openai/v1 provider: OpenAIThe user also reported that a LiteLLM bridge that previously worked now returns a
connection error, and the k8sagent pod logs showedERROR:mcp.client.streamable_http:Unexpected content type:. One user speculated the problem could be a Bedrock response header that kagent does not handle or an issue converting the Bedrock API token into SigV4 credentials, but the contributor did not confirm either theory.Another user observed that kagent currently appears to only support models that expose an OpenAI-compatible API, and that Bedrock's OpenAI endpoint seems limited to OpenAI models in us-west-2. No working Bedrock configuration appears in the thread, so for now you would need to wait for native support or contribute it.
The AWS key is referenced from a Kubernetes Secret via
apiKeySecretandapiKeySecretKey, so the key itself is not embedded in the ModelConfig YAML.Did this work for you?What @jicowan tried on GitHub (Didn't work)
Created a ModelConfig CR with provider: OpenAI, baseUrl https://bedrock-runtime.us-west-2.amazonaws.com/openai/v1 (opens external site), and model anthropic.claude-3-7-sonnet-20250219-v1:0, loading the AWS key from a Secret. Requests failed with 404 model_not_found.
Source: https://github.com/kagent-dev/kagent/issues/183#issuecomment-3560972353 (opens external site)
Did this work for you?What @jicowan tried on GitHub (Didn't work)
Re-tried the LiteLLM bridge that had previously worked for Bedrock models; the agent now returned a connection error.
Source: https://github.com/kagent-dev/kagent/issues/183#issuecomment-3560972353 (opens external site)
Did this work for you?What @jicowan tried on GitHub (Outcome unclear)
Checked the k8sagent pod logs and found ERROR:mcp.client.streamable_http:Unexpected content type:, but no fix or root cause was identified in the thread.
Source: https://github.com/kagent-dev/kagent/issues/183#issuecomment-3563746753 (opens external site)
Did this work for you?
Related questions
- kagent A2A agent invocation fails with 301 redirect on /api/a2a/kagent/<agent> path without trailing slash
- kagent 0.5.0 breaks Vertex AI project/location and Grafana MCP SSE URL in helm chart
- kagent + GPT-5: tools run but no answer, CPU busy; CancelledError: Cancelled by cancel scope — fixed?
- Can I run Claude/Codex coding agents in kagent as a declarative runtime?
- Does kagent have RBAC for OIDC users? How to restrict agent access?