Fix TypeError: messages.create(...).withResponse is not a function in Anthropic SDK

How to Fix It

Solution 1: Check for duplicate SDK versions with npm ls

This is the first diagnostic step recommended by the SDK maintainer (Source 2). Run the following command in your project root:

npm ls @anthropic-ai/sdk

If you see more than one version listed (e.g., 0.29.0 and 0.32.0), you have a duplicate. The output will show which package depends on which version. For example:

my-app@1.0.0 /path/to/project
├── @anthropic-ai/sdk@0.32.0
└─┬ traceloop@1.2.3
  └── @anthropic-ai/sdk@0.29.0

In this case, traceloop brings in its own older SDK. To fix this, you need to either:

• Update the conflicting dependency (traceloop in this example) to a version that uses the same SDK version as your project.
• Use npm overrides or resolutions (in package.json) to force the conflicting dependency to use your SDK version. For npm:

{
  "overrides": {
    "traceloop": {
      "@anthropic-ai/sdk": "^0.32.0"
    }
  }
}

For Yarn:

{
  "resolutions": {
    "@anthropic-ai/sdk": "^0.32.0"
  }
}

After adding the override, run npm install or yarn install and verify with npm ls @anthropic-ai/sdk again. You should see only one version.

Solution 2: Review OpenTelemetry or APM auto-instrumentation

If you use any OpenTelemetry, APM, or monitoring library that automatically instruments HTTP calls (e.g., traceloop, @opentelemetry/instrumentation-http, dd-trace, newrelic), check whether it wraps messages.create(). The maintainer (Source 2) specifically calls out "any OpenTelemetry/APM auto-instrumentation that might be wrapping messages.create()" as a potential cause.

To test this, temporarily disable the instrumentation library. If the error goes away, you have identified the culprit. You may need to:

• Update the instrumentation library to a version that properly handles the APIPromise type.
• Configure the instrumentation to exclude Anthropic SDK calls from wrapping.
• Or, as a workaround, avoid using .withResponse() or .asResponse() and use the raw Promise instead (see Solution 4).

Solution 3: Upgrade the Anthropic SDK to the latest version

If you are on an older version like ^0.29.0, upgrade to the current release. The GitHub issue (Source 2) reports that one user resolved the problem this way. Run:

npm install @anthropic-ai/sdk@latest

Or specify a specific recent version:

npm install @anthropic-ai/sdk@0.32.0

After upgrading, test your code. Note that this fix alone may not work if the root cause is a conflicting transitive dependency, but it eliminates the possibility that the SDK itself is the problem.

Solution 4: Workaround: avoid .withResponse() and .asResponse()

If you cannot resolve the dependency conflict immediately, you can work around the error by not calling .withResponse() or .asResponse(). The maintainer (Source 2) notes that one affected user "worked around it by avoiding .asResponse() and is no longer on the project."

Instead of:

const response = await anthropic.messages.create({...}).withResponse();

Use:

const message = await anthropic.messages.create({...});
// message is already the Message object, no need for .withResponse()

If you need the raw HTTP response headers or status code, you can use the asResponse() method only if it's available. But if you are getting this error, it's not available, so you must omit it. The MessageStream class (used in the original reporter's code) also calls .withResponse() internally, so this workaround may not apply if you are using anthropic.messages.stream(). In that case, you must fix the underlying dependency conflict.

If Nothing Works

If none of the above solutions resolve the error, the GitHub maintainer (Source 2) advises:

1. Run npm ls @anthropic-ai/sdk and verify there is exactly one version. If there is, but the error persists, check for any bundler or build tool that might be resolving modules differently in production vs. development (e.g., Webpack, esbuild, Docker layer caching).

2. Review your node_modules directory structure. In Docker, the build context may include a different set of packages than your local environment. Ensure your Dockerfile runs npm ci (clean install) rather than npm install to guarantee exact dependency resolution.

3. Open a new issue on the Anthropic SDK TypeScript GitHub repository (https://github.com/anthropics/anthropic-sdk-typescript/issues/new) with a minimal reproduction. The maintainer (Source 2) explicitly requests: "feel free to open a new issue with a minimal repro if it persists after that." Include:

   • The output of npm ls @anthropic-ai/sdk
   • Your package.json dependencies
   • A minimal code snippet that reproduces the error
   • Whether it happens in production only or also locally
   • Any OpenTelemetry or APM libraries you use

4. As a temporary workaround, you can use the raw HTTP API with curl or fetch directly, as the original reporter (Source 1) confirmed that a curl command worked fine in the production pod:

curl https://api.anthropic.com/v1/messages \
     --header "anthropic-version: 2023-06-01" \
     --header "content-type: application/json" \
     --header "x-api-key: $ANTHROPIC_API_KEY" \
     --data \
'{
  "model": "claude-3-7-sonnet-20250219",
  "messages": [{"role": "user", "content": "Hello"}],
  "max_tokens": 256,
  "stream": true
}'

This bypasses the SDK entirely and can serve as a stopgap while you debug the dependency issue.

How to Prevent It

1. Pin your SDK version and use lockfiles. Use npm ci or yarn install --frozen-lockfile in CI/CD and Docker builds to ensure the exact same dependency tree is installed every time. This prevents accidental version mismatches between environments.

2. Audit transitive dependencies. When adding any OpenTelemetry, APM, or monitoring library that auto-instruments HTTP calls, check its package.json for @anthropic-ai/sdk in its dependencies. If it lists the SDK, ensure the version range is compatible with yours. Use npm ls @anthropic-ai/sdk after installation to confirm no duplicates.

3. Use overrides proactively. If you use any instrumentation library that might bundle the Anthropic SDK, add an override in your package.json to force a single version:

{
  "overrides": {
    "@anthropic-ai/sdk": "^0.32.0"
  }
}

4. Test in production-like environments. The original reporter (Source 1) could not reproduce the error locally but saw it in Docker. Run your tests in a Docker container or CI environment that mirrors production to catch dependency conflicts early.

5. Keep the SDK updated. Regularly update @anthropic-ai/sdk to the latest version. While version alone is rarely the root cause, newer versions may have better error messages or compatibility with common instrumentation libraries.