Claude Code Setup Guide: Complete Guide for Windows, API Providers, Models and Common Errors

If you are a developer looking for a powerful AI coding assistant that can work directly from your terminal, Claude Code is worth exploring.

Unlike a traditional chatbot where you copy and paste code into a browser, Claude Code is designed to work with your project. It can inspect files, understand a codebase, make changes, run commands and help you iterate on a task from the command line.

The difficult part, however, is not always installing Claude Code.

For many developers, the real challenge starts when they want to connect Claude Code to an API provider such as Anthropic, OpenRouter or another compatible gateway.

You may encounter errors such as:

• 401 Missing Authentication
• 400 Invalid Request
• Unable to determine provider for model
• Model not found
• /v1/models returned HTTP 401
• /v1/messages returned HTTP 400

This complete Claude Code setup guide explains the configuration step by step, especially for Windows users.

What Is Claude Code?

Claude Code is Anthropic's command-line coding agent. Instead of limiting the AI to a single conversation window, it is designed to interact with your development environment and help with real programming tasks.

A typical workflow looks like this:

Your project
    ↓
Claude Code
    ↓
AI model
    ↓
Analysis + commands + code changes
    ↓
Your project

This makes it particularly useful for tasks such as debugging, refactoring, understanding an unfamiliar repository, writing tests and implementing features.

Claude Code can also be connected to compatible gateways and model providers. OpenRouter's current documentation, for example, specifically lists Claude Code as a supported coding agent and provides an Anthropic-compatible configuration path.

Why Developers Are Using Claude Code

The biggest advantage of Claude Code is its terminal-based workflow.

You can start Claude Code inside a project directory and ask it to work with the files already present in that project.

For example:

Explain this project structure.

or:

Find the reason why the login page is failing.

or:

Add validation to the registration form and run the tests.

This is different from asking a normal chatbot to generate a code snippet without access to your project.

For developers working on large repositories, that difference can be significant.

Claude Code System Requirements

Before installing Claude Code, make sure your computer has a suitable command-line environment and a working internet connection.

Windows users should also be comfortable using either Command Prompt or PowerShell because environment variables are frequently used when configuring API providers.

You should also keep your API credentials private.

Never publish an API key inside:

• GitHub repositories
• public blog posts
• screenshots
• JavaScript source code
• public configuration files

Instead, use environment variables or secure credential storage.

How to Install Claude Code on Windows

The exact installation method can change as Claude Code evolves, so always check the current official Claude Code documentation before installing a new version.

Once installed, verify that the command is available:

claude --version

If the command returns a version number, Claude Code is available from your terminal.

You can then move into your project directory:

cd C:\Projects\MyProject

and launch Claude Code:

claude

The first launch may display setup, authentication or update information.

How Claude Code Connects to AI Models

One of the most important concepts is that Claude Code is the client, while the model provider is the service that actually processes your request.

A simplified architecture looks like this:

Claude Code
     ↓
API endpoint
     ↓
Provider / Gateway
     ↓
AI model

With direct Anthropic access:

Claude Code
     ↓
api.anthropic.com
     ↓
Claude

With OpenRouter:

Claude Code
     ↓
OpenRouter
     ↓
Anthropic / Google / other providers
     ↓
Selected model

This distinction becomes important when troubleshooting authentication and model errors.

Claude Code With Anthropic API

If you use the official Anthropic API, your requests go directly to Anthropic's API infrastructure.

The authentication mechanism is different from OpenRouter's generic OpenAI-compatible authentication.

For a direct Anthropic setup, use the credentials and configuration documented by Anthropic.

For example, Windows environment variables may look conceptually like:

set ANTHROPIC_API_KEY=YOUR_ANTHROPIC_API_KEY

Do not put your real API key directly into an article or public repository.

If your Anthropic account does not have API credits, a request can still reach the Anthropic API successfully but be rejected because the account does not have sufficient API balance.

That is an important distinction:

Authentication successful
        ↓
API request accepted
        ↓
Billing check
        ↓
Insufficient credits

An insufficient-credit error is therefore different from an invalid API-key error.

Claude Code With OpenRouter

OpenRouter provides a unified API gateway for many AI models and providers.

Its current documentation specifically describes Claude Code integration using environment variables and an Anthropic-compatible endpoint.

A typical OpenRouter setup uses an OpenRouter API key and an OpenRouter base URL.

The current OpenRouter documentation shows:

ANTHROPIC_BASE_URL=https://openrouter.ai/api
ANTHROPIC_AUTH_TOKEN=<openrouter-api-key>

along with model-related configuration.

The exact configuration can change with Claude Code releases, so check the current OpenRouter and Claude Code documentation before deploying a production configuration.

OpenRouter Environment Variables

On Windows Command Prompt, the basic idea is:

set ANTHROPIC_BASE_URL=https://openrouter.ai/api
set ANTHROPIC_AUTH_TOKEN=YOUR_OPENROUTER_KEY

Then configure the model you want to use.

OpenRouter uses a provider/model naming structure. For example:

anthropic/claude-sonnet-4

The provider prefix is important because it tells the gateway which provider/model combination you are requesting. OpenRouter's current documentation explicitly describes model identifiers using this format.

Choosing an OpenRouter Model

Do not assume that every model works identically with Claude Code.

When choosing a model, check:

• coding capability
• context window
• tool support
• vision support
• reasoning capability
• latency
• price
• provider availability

OpenRouter currently provides model discovery and model information through its catalog, and its newer MCP integration can expose live model and pricing information to supported agents.

Claude Code With a Custom API Gateway

Advanced users may route Claude Code through a local or remote gateway.

For example:

Claude Code
      ↓
http://localhost:20128
      ↓
Gateway
      ↓
Provider
      ↓
Model

This can be useful when a gateway provides:

• multiple model providers
• model routing
• fallback
• centralized credentials
• local development endpoints
• provider-specific configuration

However, this is also where many configuration errors occur.

Your gateway must understand the model identifier you provide.

For example:

antigravity/claude-sonnet-4-6

contains both:

provider = antigravity
model    = claude-sonnet-4-6

while:

claude-sonnet-4-6

does not necessarily tell the gateway which provider should process the request.

Important Claude Code Environment Variables

Three environment variables are particularly important when working with custom configurations.

ANTHROPIC_BASE_URL

This specifies the API endpoint.

For direct Anthropic access, the endpoint is Anthropic's API.

For a gateway, it may instead look like:

set ANTHROPIC_BASE_URL=http://localhost:20128

The exact value depends on the provider you are using.

ANTHROPIC_AUTH_TOKEN

This provides the authentication credential expected by the configured endpoint.

For example:

set ANTHROPIC_AUTH_TOKEN=YOUR_TOKEN

Do not assume that an Anthropic API key, OpenRouter API key and local gateway token are interchangeable.

They may use completely different authentication systems.

ANTHROPIC_MODEL

This specifies the model Claude Code should request.

For example:

set ANTHROPIC_MODEL=anthropic/claude-sonnet-4

when using a gateway that expects provider/model notation.

The correct model identifier depends on the gateway's model catalog.

How to Change the Claude Code Model

On Windows CMD:

set ANTHROPIC_MODEL=YOUR_PROVIDER/YOUR_MODEL

Then start Claude Code:

claude

For example:

set ANTHROPIC_MODEL=anthropic/claude-sonnet-4
claude

If your gateway exposes a different model identifier, use exactly the identifier returned by its /v1/models endpoint.

You can test the endpoint with:

curl http://localhost:20128/v1/models

The response should show the model identifiers available through your gateway.

Common Claude Code 401 Errors

A 401 normally indicates an authentication problem.

For example:

401 Missing Authentication header

means the server did not receive the authentication information it expected.

Common causes include:

• missing API key
• wrong environment variable
• incorrect authentication type
• expired credential
• wrong endpoint
• gateway not forwarding authentication
• using an OpenRouter credential against the wrong endpoint

If the provider expects:

Authorization: Bearer ...

but your client sends another authentication mechanism, the request can fail.

Likewise, if a provider expects a provider-specific API-key header, a generic Bearer configuration may not work.

Always check the provider's current API documentation.

Common Claude Code 400 Errors

A 400 means the server received the request but rejected something about it.

One common example is:

The request body is not valid JSON

This can happen when manually testing an API from Windows CMD or PowerShell and incorrectly escaping the JSON body.

Another common reason is an invalid model name.

For example:

Unable to determine provider for model

usually means the gateway cannot map the supplied model identifier to a provider.

Model Provider Errors

Suppose your gateway returns:

Unable to determine provider for model
'antigravity-only'

The problem may be the model identifier rather than the API key.

If the gateway expects:

antigravity/claude-sonnet-4-6

then:

antigravity-only

may not be sufficient.

Always inspect:

/v1/models

before guessing model names.

This is one of the simplest troubleshooting techniques for multi-provider AI gateways.

How to Test Your API Connection

Before blaming Claude Code, test the API endpoint separately.

For a local gateway:

curl http://localhost:20128/v1/models

If the endpoint works, you should receive a model list.

Then test the actual inference endpoint using the authentication and request format required by that gateway.

This creates a useful troubleshooting sequence:

1. Is the gateway running?
          ↓
2. Does /v1/models respond?
          ↓
3. Is authentication accepted?
          ↓
4. Does the selected model exist?
          ↓
5. Does /v1/messages work?
          ↓
6. Does Claude Code work?

This approach is much faster than changing five configuration settings at the same time.

Claude Code Security Best Practices

AI coding tools can access your project files and execute commands, so security deserves serious attention.

Never share credentials such as:

sk-ant-...
sk-or-...

in public repositories or screenshots.

Use environment variables:

set ANTHROPIC_AUTH_TOKEN=YOUR_TOKEN

and keep secrets outside source code.

Also review what permissions your coding agent has before allowing it to modify production systems.

For development projects, consider:

• Git version control
• regular backups
• isolated test environments
• restricted API keys
• spending limits
• separate development credentials

When using an API gateway, also understand where requests are routed and what data policies apply.

Claude Code vs Other AI Coding Tools

Claude Code is only one option in the rapidly growing AI coding-agent ecosystem.

Developers may also consider:

• Cursor
• GitHub Copilot
• Gemini CLI
• Codex CLI
• OpenCode
• Cline
• other agentic coding tools

The best option depends on your workflow.

If you prefer a terminal-first development experience, Claude Code is particularly interesting.

If you prefer an integrated editor, an IDE-based tool may be more convenient.

If you want access to multiple providers through one API gateway, OpenRouter can be attractive because it allows supported coding agents to use a unified endpoint and switch between model identifiers.

Frequently Asked Questions

Is Claude Code free?

Claude Code itself is a software tool, but the AI inference behind it may involve a subscription or API usage depending on how you authenticate and which provider you use.

Always check the current pricing and terms of your chosen provider.

Can I use Claude Code with OpenRouter?

Yes. OpenRouter currently documents Claude Code as a supported coding agent and provides an Anthropic-compatible configuration approach.

Can I use a custom API gateway with Claude Code?

Yes, provided that the gateway exposes a compatible API and correctly handles authentication, model routing and the required request format.

Why does Claude Code say it cannot determine the provider?

Usually, the model identifier does not contain enough information for the gateway to identify the provider, or the model has not been configured in the gateway.

Check:

curl http://localhost:20128/v1/models

and use an identifier actually returned by the gateway.

Why am I getting a 401 error?

Check your API key, authentication method, endpoint and environment variables.

Also make sure you are not mixing credentials between Anthropic, OpenRouter and a custom gateway.

Why am I getting a 400 error?

A 400 response can have many causes. Check the exact server message.

Invalid JSON, an invalid model, unsupported parameters and incorrect request formatting are common causes.

Conclusion

Claude Code can become a powerful part of a developer's workflow, but getting the configuration right is important.

The easiest way to avoid frustration is to think about the system as several separate layers:

Claude Code
     ↓
Base URL
     ↓
Authentication
     ↓
Provider
     ↓
Model
     ↓
Inference

If something fails, identify which layer is responsible instead of changing everything at once.

For a direct Anthropic setup, verify your Anthropic credentials and account status.

For OpenRouter, verify the OpenRouter endpoint, authentication and provider/model identifier. OpenRouter currently supports Claude Code and provides specific configuration guidance for it.

For a custom gateway, first check /v1/models, then confirm the exact model identifier and authentication method.

Once those pieces are working, Claude Code becomes much easier to use—and troubleshooting becomes far less mysterious.