Calling Claude on AWS Bedrock with OpenAI SDK

Calling Claude on AWS Bedrock with OpenAI SDK

Overview

AWS Bedrock hosts foundation models from multiple providers, including Anthropic’s Claude. If you already have AWS_REGION, AWS_BEARER_TOKEN_BEDROCK, and CLAUDE_CODE_USE_BEDROCK configured, this guide explains how to call Claude through the OpenAI-compatible interface and clarifies what each variable does.

Environment Variables

Variable	Purpose
AWS_REGION	Bedrock service region (e.g. us-east-2)
AWS_BEARER_TOKEN_BEDROCK	Bearer token for Bedrock authentication
CLAUDE_CODE_USE_BEDROCK	Claude Code CLI only — tells the claude CLI tool to route through Bedrock instead of api.anthropic.com

CLAUDE_CODE_USE_BEDROCK is exclusively used by the Claude Code CLI agent. It has no effect when you call models programmatically with the OpenAI SDK.

Two Bedrock Endpoints

AWS Bedrock exposes two distinct endpoint styles. Understanding the difference is critical.

bedrock-mantle.*.api.aws — OpenAI-Compatible

This is the OpenAI-compatible endpoint. It accepts the standard /v1/chat/completions format and uses Bearer Token authentication.

https://bedrock-mantle.us-east-2.api.aws/v1

• Auth: Authorization: Bearer <token>
• Request format: Same as OpenAI ({"model": "...", "messages": [...]})
• Compatible with OpenAI SDK: Yes

bedrock-runtime.*.amazonaws.com — Native Bedrock

This is the native Bedrock endpoint. It uses a completely different API format and requires AWS Signature V4 authentication.

https://bedrock-runtime.us-east-2.amazonaws.com

• Auth: AWS SigV4 (HMAC-based request signing with AWS_ACCESS_KEY_ID + AWS_SECRET_ACCESS_KEY)
• Request format: POST /model/{modelId}/invoke with provider-specific body
• Compatible with OpenAI SDK: No

Feature	bedrock-mantle	bedrock-runtime
Auth method	Bearer Token	AWS SigV4 Signature
API format	OpenAI-compatible	Bedrock-native
OpenAI SDK works?	Yes	No
SDK required	openai	boto3

Calling Claude with OpenAI SDK

Since bedrock-mantle is OpenAI-compatible, you can use the OpenAI Python SDK directly:

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://bedrock-mantle.us-east-2.api.aws/v1",
    api_key=os.environ["AWS_BEARER_TOKEN_BEDROCK"],
)

response = client.chat.completions.create(
    model="anthropic.claude-sonnet-4-20250514-v1:0",
    messages=[{"role": "user", "content": "Hello!"}],
)
print(response.choices[0].message.content)

Listing Available Models

Not all models may be available in your region or account. Query the endpoint to check:

models = client.models.list()
for m in models.data:
    print(m.id)

Or via curl:

curl -H "Authorization: Bearer $AWS_BEARER_TOKEN_BEDROCK" \
     https://bedrock-mantle.us-east-2.api.aws/v1/models

If Claude models are missing from the list, possible causes:

1. Region availability — Claude may not be enabled in us-east-2. Common regions: us-east-1, us-west-2.
2. Model access not granted — In the AWS Console → Bedrock → Model access, you must explicitly request access to Anthropic Claude models.
3. Endpoint limitations — The bedrock-mantle endpoint may only expose a subset of models.

Calling Claude with boto3 (Native Endpoint)

If you need to use the native bedrock-runtime endpoint (e.g. when bedrock-mantle doesn’t list Claude), use boto3:

import boto3
import json

client = boto3.client("bedrock-runtime", region_name="us-east-1")

response = client.invoke_model(
    modelId="anthropic.claude-sonnet-4-20250514-v1:0",
    contentType="application/json",
    body=json.dumps({
        "anthropic_version": "bedrock-2023-05-31",
        "messages": [{"role": "user", "content": "Hello!"}],
        "max_tokens": 1024,
    }),
)
result = json.loads(response["body"].read())
print(result)

This requires IAM credentials (AWS_ACCESS_KEY_ID / AWS_SECRET_ACCESS_KEY) rather than a Bearer token.

Using LiteLLM as a Proxy

LiteLLM can translate OpenAI-format requests to Bedrock-native calls, acting as a local proxy:

# Start the proxy
litellm --model bedrock/anthropic.claude-sonnet-4-20250514-v1:0

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:4000/v1",
    api_key="sk-anything",  # any value works
)

response = client.chat.completions.create(
    model="bedrock/anthropic.claude-sonnet-4-20250514-v1:0",
    messages=[{"role": "user", "content": "Hello!"}],
)
print(response.choices[0].message.content)

LiteLLM reads AWS_REGION and AWS credentials from environment variables automatically and handles SigV4 signing internally.

Summary

┌─────────────────┐     Bearer Token      ┌──────────────────────────────────┐
│   OpenAI SDK    │ ───────────────────▶   │  bedrock-mantle.*.api.aws/v1   │
└─────────────────┘                        └──────────────────────────────────┘

┌─────────────────┐     SigV4 Signing      ┌──────────────────────────────────┐
│     boto3       │ ───────────────────▶   │  bedrock-runtime.*.amazonaws.com │
└─────────────────┘                        └──────────────────────────────────┘

┌─────────────────┐     OpenAI format      ┌─────────┐   SigV4   ┌──────────┐
│   OpenAI SDK    │ ───────────────────▶   │ LiteLLM │ ────────▶ │ Bedrock  │
└─────────────────┘                        └─────────┘           └──────────┘

• Have a Bearer Token? → Use bedrock-mantle with OpenAI SDK directly.
• Have IAM credentials? → Use boto3 or LiteLLM as a proxy.
• Using Claude Code CLI? → Set CLAUDE_CODE_USE_BEDROCK=1 and it handles everything.