How to Handle Multi-Language Inputs Correctly in Claude Workflows with n8n

Open your Webhook node and ensure the response content-type is set to application/json with an explicit charset. By default, n8n webhooks accept UTF-8, but downstream nodes may misinterpret the encoding if the header is ambiguous. In the Webhook node settings, set Response Content-Type to 'application/json; charset=utf-8'. Also set the HTTP Method to POST so the body is parsed as JSON rather than URL-encoded, which can mangle multi-byte characters.

Different systems may send the same character in different Unicode forms (NFC vs NFD). For example, an accented 'é' can be one codepoint or two. Claude handles both, but downstream string operations in n8n (like truncation or length checks) can split multi-byte sequences if the form is inconsistent. Add a Code node right after your Webhook to normalize all text input to NFC form. This ensures consistent character counting and prevents truncation bugs when you later check token limits.

1// Code node — JavaScript
2const items = $input.all();
3
4for (const item of items) {
5  const userText = item.json.message || item.json.text || '';
6  // Normalize to NFC (composed form)
7  item.json.normalizedText = userText.normalize('NFC');
8  // Accurate character count for multi-byte text
9  item.json.charCount = [...item.json.normalizedText].length;
10}
11
12return items;

In your Anthropic node (or AI Agent node with Claude as the model), reference the normalizedText field from the previous Code node. Use the expression {{ $json.normalizedText }} in the user message field. Do not use string concatenation with + in expressions for multi-language text — always use template literals or direct field references. String concatenation in n8n expressions can sometimes trigger implicit toString() calls that re-encode characters.

If you are calling the Anthropic API directly via an HTTP Request node instead of the built-in Anthropic node, you must set the Content-Type header to 'application/json; charset=utf-8' explicitly. In the HTTP Request node, go to Options → Headers and add this header. Also ensure the body is sent as JSON (not raw text), because raw text mode may apply platform-default encoding which varies between Linux and Windows n8n installations.

1// HTTP Request node configuration (set in node UI):
2// Method: POST
3// URL: https://api.anthropic.com/v1/messages
4// Headers:
5//   Content-Type: application/json; charset=utf-8
6//   x-api-key: {{ $credentials.anthropicApi.apiKey }}
7//   anthropic-version: 2023-06-01
8// Body (JSON):
9{
10  "model": "claude-3-5-sonnet-20241022",
11  "max_tokens": 1024,
12  "messages": [
13    {
14      "role": "user",
15      "content": "{{ $json.normalizedText }}"
16    }
17  ]
18}

Add a second Code node after the Claude response to verify that the response text is valid UTF-8 and contains the expected characters. This is especially important for CJK (Chinese, Japanese, Korean) text and RTL (right-to-left) languages like Arabic and Hebrew. Log any encoding anomalies so you can catch issues before they reach the end user. This validation step also protects against rare cases where the API response is double-encoded.

1// Code node — JavaScript
2const items = $input.all();
3
4for (const item of items) {
5  const responseText = item.json.message?.content?.[0]?.text
6    || item.json.output
7    || '';
8
9  // Check for common encoding corruption patterns
10  const hasMojibake = /Ã[\x80-\xBF]/.test(responseText);
11  const hasReplacementChar = responseText.includes('\uFFFD');
12
13  item.json.responseText = responseText;
14  item.json.encodingValid = !hasMojibake && !hasReplacementChar;
15
16  if (!item.json.encodingValid) {
17    item.json.encodingWarning = 'Response contains encoding artifacts — check upstream headers';
18  }
19}
20
21return items;

If your workflow responds via the Webhook node (Respond to Webhook), ensure the response also uses UTF-8. In the Respond to Webhook node, set the Response Content-Type to 'application/json; charset=utf-8'. If you are returning HTML (for a chat widget), use 'text/html; charset=utf-8'. Without this, browsers may render the response in ISO-8859-1, turning CJK and accented characters into garbage. This closes the encoding loop end-to-end.