Claude API Base URL Configuration Guide for Cursor, Cline, Dify, Open WebUI, and More

Base URL mistakes are one of the most common reasons Claude API integrations fail.

The API key can be correct. The model name can be correct. But if the tool expects an OpenAI-compatible base URL and you forget /v1, the request may never reach the right endpoint. In the opposite direction, if a tool expects an Anthropic-style base host and you add /v1 in the base field, the client may generate a broken path.

This guide explains which ClaudeAPI URL to use in Cursor, Cline, Dify, ChatBox, Cherry Studio, Open WebUI, and Anthropic-compatible clients.

The short rule

Use this table first. Most configuration problems come from mixing up these two URL styles.

The practical rule:

OpenAI-compatible tool  -> https://gw.claudeapi.com/v1
Anthropic-compatible tool -> https://gw.claudeapi.com

OpenAI-compatible tool  -> https://gw.claudeapi.com/v1
Anthropic-compatible tool -> https://gw.claudeapi.com

Also avoid a trailing slash:

Correct: https://gw.claudeapi.com/v1
Avoid:   https://gw.claudeapi.com/v1/

Correct: https://gw.claudeapi.com/v1
Avoid:   https://gw.claudeapi.com/v1/

Some tools handle the trailing slash gracefully. Others concatenate paths poorly and produce double slashes or 404 errors. Keeping the URL clean saves time.

Base URL vs endpoint

A base URL is not the full API endpoint.

For OpenAI-compatible clients, you usually enter:

https://gw.claudeapi.com/v1

https://gw.claudeapi.com/v1

Then the SDK or tool appends the actual endpoint:

/chat/completions
/models

/chat/completions
/models

So the final request becomes:

https://gw.claudeapi.com/v1/chat/completions

https://gw.claudeapi.com/v1/chat/completions

Do not paste the full endpoint into a field named “Base URL”, “API Base”, or “API Host”:

Wrong for a Base URL field:
https://gw.claudeapi.com/v1/chat/completions

Wrong for a Base URL field:
https://gw.claudeapi.com/v1/chat/completions

That is the full request path, not the base.

Cursor

Use the OpenAI-compatible provider option.

Typical setup:

1. Open Cursor settings.
2. Go to the models or API provider section.
3. Select OpenAI Compatible.
4. Enter https://gw.claudeapi.com/v1 as the base URL.
5. Paste your ClaudeAPI key.
6. Enter or select a Claude model.
7. Run the built-in verification if Cursor shows one.

Successful configuration usually means the verify button succeeds or chat/completion requests return normally.

Common issues:

Cline

Cline also works through an OpenAI-compatible configuration.

Recommended steps:

1. Open the Cline panel in VS Code.
2. Open settings.
3. Choose OpenAI Compatible as the provider.
4. Set Base URL to https://gw.claudeapi.com/v1.
5. Paste your ClaudeAPI key.
6. Manually enter the full model ID.
7. Save and send a short test message.

If Cline still shows an old model after saving, reload VS Code and check the provider settings again.

Dify

In Dify, configure ClaudeAPI as an OpenAI-compatible model provider.

Important: Dify’s API Base URL field should contain only the base path:

https://gw.claudeapi.com/v1

https://gw.claudeapi.com/v1

Do not enter:

https://gw.claudeapi.com/v1/chat/completions

https://gw.claudeapi.com/v1/chat/completions

Dify will append the actual endpoint internally.

After saving the provider, add the model name in model management if Dify does not automatically list it. A successful setup means the model appears as available and a Dify app can send a test message with a normal response.

ChatBox

ChatBox can expose more than one provider mode. When using a ClaudeAPI key, choose the OpenAI-compatible mode rather than the native Claude/Anthropic mode.

If you choose the native Claude API mode with a ClaudeAPI OpenAI-compatible key, authentication or endpoint errors are likely.

Cherry Studio

Use the OpenAI Compatible provider type.

Do not select Anthropic unless you are intentionally configuring an Anthropic-native endpoint supported by your environment.

If the model dropdown is empty, add a model manually with the exact ID from the ClaudeAPI console.

Open WebUI

Open WebUI can connect to OpenAI-compatible endpoints through its admin settings or environment configuration, depending on how it is deployed.

Use:

After saving, test the model list:

curl https://gw.claudeapi.com/v1/models \
  -H "Authorization: Bearer YOUR_CLAUDEAPI_KEY"

curl https://gw.claudeapi.com/v1/models \
  -H "Authorization: Bearer YOUR_CLAUDEAPI_KEY"

A successful response should look like an OpenAI-style model list:

{
  "object": "list",
  "data": [
    {
      "id": "claude-haiku-4-5-20251001",
      "object": "model"
    }
  ]
}

{
  "object": "list",
  "data": [
    {
      "id": "claude-haiku-4-5-20251001",
      "object": "model"
    }
  ]
}

If Open WebUI shows an empty model list, check:

• the base URL includes /v1
• the API key was saved in the admin panel
• the account has models enabled
• you are configuring as an admin user

Anthropic-compatible clients and Claude Desktop notes

Anthropic-style clients usually work differently from OpenAI-compatible clients. They call endpoints such as:

/v1/messages

/v1/messages

That means the base host should usually be:

https://gw.claudeapi.com

https://gw.claudeapi.com

Then the client appends:

/v1/messages

/v1/messages

So the final request becomes:

https://gw.claudeapi.com/v1/messages

https://gw.claudeapi.com/v1/messages

Do not put /v1 in the base field if the Anthropic client already appends /v1/messages.

For Claude Desktop specifically, custom third-party API endpoint support can vary by version and distribution. If your version exposes a custom API base URL or Anthropic-compatible endpoint setting, use:

https://gw.claudeapi.com

https://gw.claudeapi.com

If your Claude Desktop build only supports the official Claude account flow and MCP server configuration, it may not support replacing the inference endpoint directly. In that case, use ClaudeAPI with tools that explicitly support OpenAI-compatible or Anthropic-compatible custom endpoints, such as Cursor, Cline, Dify, ChatBox, Cherry Studio, or Open WebUI.

Quick troubleshooting flow

When a tool fails, test the API outside the tool first. This separates account/network problems from UI configuration problems.

Test OpenAI-compatible chat completions

Use this for Cursor, Cline, Dify, ChatBox, Cherry Studio, and Open WebUI:

curl https://gw.claudeapi.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_CLAUDEAPI_KEY" \
  -d '{
    "model": "claude-haiku-4-5-20251001",
    "messages": [{"role": "user", "content": "Reply with OK only."}],
    "max_tokens": 16
  }'

curl https://gw.claudeapi.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_CLAUDEAPI_KEY" \
  -d '{
    "model": "claude-haiku-4-5-20251001",
    "messages": [{"role": "user", "content": "Reply with OK only."}],
    "max_tokens": 16
  }'

Expected result:

• the response contains a choices array
• choices[0].message.content contains text
• usage may show token counts

Test Anthropic-style messages

Use this only when you are testing an Anthropic-compatible path:

curl https://gw.claudeapi.com/v1/messages \
  -H "Content-Type: application/json" \
  -H "x-api-key: YOUR_CLAUDEAPI_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -d '{
    "model": "claude-haiku-4-5-20251001",
    "max_tokens": 100,
    "messages": [{"role": "user", "content": "Reply with OK only."}]
  }'

curl https://gw.claudeapi.com/v1/messages \
  -H "Content-Type: application/json" \
  -H "x-api-key: YOUR_CLAUDEAPI_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -d '{
    "model": "claude-haiku-4-5-20251001",
    "max_tokens": 100,
    "messages": [{"role": "user", "content": "Reply with OK only."}]
  }'

Expected result:

• the response contains a content array
• content[0].text contains the model reply

Error reference

FAQ

Should the Base URL end with a slash?

No. Use:

https://gw.claudeapi.com/v1

https://gw.claudeapi.com/v1

not:

https://gw.claudeapi.com/v1/

https://gw.claudeapi.com/v1/

Why do OpenAI-compatible tools need /v1?

OpenAI-compatible SDKs expect the base URL to include the API version. They then append paths such as /chat/completions or /models.

Why do Anthropic-compatible clients use the host without /v1?

Anthropic-style clients often append /v1/messages themselves. If you put /v1 in the base field, the final path can become duplicated or malformed.

Is the API key the same as an Anthropic official key?

No. Use the key generated in the ClaudeAPI console. Keys from other providers are not interchangeable.

Can I use the same ClaudeAPI key in multiple tools?

Yes. You can use the same key in multiple clients, and usage will be billed according to the requests each client sends. For team environments, separate keys per tool or environment make auditing easier.

Where do I find model names?

Check the ClaudeAPI model page or console and copy the exact model ID. Common examples include:

claude-haiku-4-5-20251001
claude-sonnet-4-6
claude-opus-4-8

claude-haiku-4-5-20251001
claude-sonnet-4-6
claude-opus-4-8

Final checklist

Before debugging anything else, verify these five items:

• [ ] You selected OpenAI Compatible mode for OpenAI-style tools.
• [ ] The Base URL is https://gw.claudeapi.com/v1 for OpenAI-compatible tools.
• [ ] The Base URL is https://gw.claudeapi.com only for Anthropic-compatible base-host fields.
• [ ] The key comes from ClaudeAPI and has no leading or trailing spaces.
• [ ] The model ID is copied exactly from the console.

Once you remember the rule, new tools are easier to configure: first identify whether the tool speaks OpenAI-compatible or Anthropic-compatible API, then choose the matching base URL.

Related guides

• Migrate from OpenAI API to Claude API
• Continue VS Code Claude API setup guide
• Claude API pricing and model selection guide
• Claude API cost estimation and budget guide

Sources

• Cursor documentation
• Cline documentation
• Dify documentation
• Open WebUI documentation
• Anthropic Messages API documentation