I tried calling Opus 4.7 from LangChain using Bedrock's Anthropic-compatible endpoint (Mantle)

The Claude Opus 4.7 released today was introduced in the AWS official blog with a client called AnthropicBedrockMantle.

from anthropic import AnthropicBedrockMantle
client = AnthropicBedrockMantle(aws_region=REGION)

The Mantle documentation only mentions OpenAI-compatible APIs (Chat Completions / Responses). I was curious about where this client connects to, so I investigated.

Identifying the endpoint from debug logs

I enabled debug logging with AnthropicBedrockMantle and made a call.

pip install "anthropic[bedrock]"

import logging
logging.basicConfig(level=logging.DEBUG)

from anthropic import AnthropicBedrockMantle

client = AnthropicBedrockMantle(aws_region="ap-northeast-1")

response = client.messages.create(
    model="anthropic.claude-opus-4-7",
    max_tokens=256,
    messages=[{"role": "user", "content": "Hello!"}]
)

print(response.content[0].text)

It worked fine, and the logs output the following lines:

DEBUG:anthropic._base_client:Sending HTTP Request: POST https://bedrock-mantle.ap-northeast-1.api.aws/anthropic/v1/messages
INFO:httpx:HTTP Request: POST https://bedrock-mantle.ap-northeast-1.api.aws/anthropic/v1/messages "HTTP/1.1 200 OK"

I found that the bedrock-mantle endpoint has a path /anthropic/v1/messages. This is an Anthropic Messages API-compatible endpoint, different from the OpenAI-compatible paths (/v1/chat/completions, /v1/responses) mentioned in the Mantle documentation.

Path compatibility status

Here are the results of testing each path with Opus 4.7:

Currently, only Opus 4.7 supports this Anthropic-compatible path. Opus 4.6 and Sonnet 4.6 do not work with it.

Testing directly with curl

Now that I know the endpoint, I confirmed it with curl. The SigV4 signature service name is bedrock-mantle.

curl -s -X POST \
  "https://bedrock-mantle.ap-northeast-1.api.aws/anthropic/v1/messages" \
  --aws-sigv4 "aws:amz:ap-northeast-1:bedrock-mantle" \
  --user "${AWS_ACCESS_KEY_ID}:${AWS_SECRET_ACCESS_KEY}" \
  -H "x-amz-security-token: ${AWS_SESSION_TOKEN}" \
  -H "Content-Type: application/json" \
  -H "anthropic-version: 2023-06-01" \
  -d '{
    "model": "anthropic.claude-opus-4-7",
    "max_tokens": 64,
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

{
  "model": "claude-opus-4-7",
  "content": [
    {
      "type": "text",
      "text": "Hello! How can I help you today?"
    }
  ],
  "stop_reason": "end_turn",
  "usage": {
    "input_tokens": 24,
    "output_tokens": 15,
    "service_tier": "standard"
  }
}

Note that the anthropic-version header is required.

Using plain Anthropic SDK with short-term Bedrock API keys

Here's the main point: since the endpoint is Anthropic Messages API-compatible, I tried using the plain Anthropic client by replacing the base_url.

For authentication, I used a short-term Bedrock API key. These can be issued with aws-bedrock-token-generator and are valid for up to 12 hours. You don't need to create an IAM user; keys can be generated from your current credentials.

pip install aws-bedrock-token-generator

from aws_bedrock_token_generator import provide_token
import anthropic

token = provide_token(region='ap-northeast-1')

client = anthropic.Anthropic(
    base_url="https://bedrock-mantle.ap-northeast-1.api.aws/anthropic",
    api_key=token
)

response = client.messages.create(
    model="anthropic.claude-opus-4-7",
    max_tokens=64,
    messages=[{"role": "user", "content": "Hello!"}]
)

print(response.content[0].text)
print(f"model: {response.model}")
print(f"usage: input={response.usage.input_tokens}, output={response.usage.output_tokens}")

Hello! How can I help you today?
model: claude-opus-4-7
usage: input=24, output=15

It worked! I was able to call Opus 4.7 through Bedrock using the plain Anthropic SDK with the same user experience as the Anthropic API. The key points are:

• Set base_url to https://bedrock-mantle.{region}.api.aws/anthropic
• Pass the short-term Bedrock API key to api_key
• Everything else works the same as the normal Anthropic API

This opens the possibility of using existing code and tools built for the Anthropic SDK within Bedrock's secure environment.

Confirmed working with LangChain

It also works with ChatAnthropic from langchain-anthropic by simply replacing the anthropic_api_url.

pip install langchain-anthropic

from aws_bedrock_token_generator import provide_token
from langchain_anthropic import ChatAnthropic

token = provide_token(region='ap-northeast-1')

llm = ChatAnthropic(
    model="anthropic.claude-opus-4-7",
    anthropic_api_key=token,
    anthropic_api_url="https://bedrock-mantle.ap-northeast-1.api.aws/anthropic",
    max_tokens=64,
)

response = llm.invoke("Amazon Bedrockとは？1行で。")
print(response.content)

Amazon Bedrockとは、AWSが提供する、複数の基盤モデル（FM）をAPI経由で利用できるフルマネージド型の生成AIサービスです。

I confirmed that ecosystems based on the Anthropic SDK can use Bedrock models just by replacing the endpoint.

Conclusion

Bedrock Mantle has added an Anthropic Messages API-compatible endpoint (/anthropic/v1/messages).

Until now, Mantle has only provided OpenAI-compatible APIs (Chat Completions / Responses), but in February 2026, a strategic partnership between OpenAI and Amazon was announced, which might explain this expansion toward bundling native APIs from various vendors.

Currently, there are some limitations:

• Only supports Opus 4.7 (Opus 4.6 / Sonnet 4.6 are not supported)
• The /anthropic/v1/messages path is not yet documented in official documentation

While it may be too early for production use, the ability to seamlessly call Bedrock models from workflows using the Anthropic SDK is very attractive. I look forward to future developments.

Reference Links

• AWS Official Blog - Introducing Anthropic's Claude Opus 4.7 model in Amazon Bedrock
• Mantle Documentation - Generate responses using OpenAI APIs
• Bedrock API Key Generation
• Strategic Partnership between OpenAI and Amazon