> ## Documentation Index
> Fetch the complete documentation index at: https://docs-apexspriteai.reliatrack.org/llms.txt
> Use this file to discover all available pages before exploring further.

# Configure Claude Code: config.json and environment variables

> Set up your shell environment and Claude Code configuration file to connect ApexSpriteAI components. Covers ANTHROPIC_BASE_URL, config.json, and MCP settings.

Claude Code reads a JSON configuration file at startup to decide where to send API requests and how to behave. By default it connects to Anthropic's cloud API, which requires an active Anthropic subscription. ApexSpriteAI overrides this default by pointing Claude Code at your local LM Studio server instead. This page explains the configuration file, the environment variable alternative, and where MCP settings are stored.

## The `~/.claude/config.json` file

Claude Code stores its configuration in `~/.claude/config.json` on your Mac. If the file does not exist yet, create it — Claude Code reads it automatically on each invocation.

The most important field is `ANTHROPIC_BASE_URL`. When this field is present, Claude Code bypasses Anthropic's OAuth flow entirely and routes all `/v1/messages` requests to the URL you specify. This is how ApexSpriteAI directs traffic to LM Studio.

### Example configuration for remote GPU (Tailscale)

```json theme={null}
{
  "env": {
    "ANTHROPIC_BASE_URL": "http://100.82.56.40:1234"
  }
}
```

Replace `100.82.56.40` with your GPU server's Tailscale IP address.

### Example configuration for local-only mode

```json theme={null}
{
  "env": {
    "ANTHROPIC_BASE_URL": "http://localhost:1234"
  }
}
```

<Note>
  Do not include a trailing slash or the `/v1/messages` path in `ANTHROPIC_BASE_URL`. Claude Code appends the path automatically.
</Note>

### Full config.json with MCP settings

If you have MCP servers configured, your file will contain an additional `mcpServers` key alongside the `env` block:

```json theme={null}
{
  "env": {
    "ANTHROPIC_BASE_URL": "http://100.82.56.40:1234"
  },
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/yourname/projects"]
    },
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_..."
      }
    }
  }
}
```

<Warning>
  `~/.claude/config.json` may contain API tokens for MCP servers. Do not commit this file to version control. Add `~/.claude/config.json` to your global `.gitignore`.
</Warning>

## Setting `ANTHROPIC_BASE_URL` as a shell environment variable

As an alternative to `config.json`, you can export `ANTHROPIC_BASE_URL` in your shell profile. This is useful when you want to switch between local and cloud mode without editing a file, or when running Claude Code in CI pipelines.

Add one of the following to your `~/.zshrc` or `~/.bash_profile`:

<CodeGroup>
  ```bash Tailscale remote GPU theme={null}
  export ANTHROPIC_BASE_URL="http://100.82.56.40:1234"
  ```

  ```bash Local-only theme={null}
  export ANTHROPIC_BASE_URL="http://localhost:1234"
  ```
</CodeGroup>

Reload your shell after saving:

```bash theme={null}
source ~/.zshrc
```

<Note>
  When both `config.json` and a shell environment variable are present, the shell environment variable takes precedence.
</Note>

## MCP configuration

MCP (Model Context Protocol) servers are registered with the `claude mcp add` command. Claude Code writes the resulting configuration into `~/.claude/config.json` under the `mcpServers` key automatically — you do not need to edit the file by hand when adding a new MCP server.

```bash theme={null}
claude mcp add filesystem npx -y @modelcontextprotocol/server-filesystem /path/to/projects
```

After running this command, verify the entry was written:

```bash theme={null}
cat ~/.claude/config.json
```

The `mcpServers` block should contain a new entry for `filesystem`.

<Tip>
  Because MCP tools execute locally on your Mac — not on the remote GPU server — they work regardless of whether you are in local-only or remote GPU mode. The model on the remote server returns a JSON payload instructing Claude Code which tool to call, and Claude Code executes the tool locally.
</Tip>

## Validating your configuration

Run the following test to confirm that Claude Code is reaching LM Studio and not Anthropic's cloud:

<Steps>
  <Step title="Check the environment variable is set">
    ```bash theme={null}
    echo $ANTHROPIC_BASE_URL
    ```

    Expected output (remote GPU example):

    ```
    http://100.82.56.40:1234
    ```

    If this is empty, your shell profile change has not been reloaded. Run `source ~/.zshrc` and check again, or verify that `config.json` contains the correct value.
  </Step>

  <Step title="Confirm LM Studio is reachable">
    ```bash theme={null}
    curl -s http://<your-tailscale-ip>:1234/v1/models | python3 -m json.tool
    ```

    You should receive a JSON object listing the currently loaded model. A connection refused error means LM Studio is not running or the bind address is not set to `0.0.0.0`.
  </Step>

  <Step title="Run a quick Claude Code session">
    ```bash theme={null}
    claude "What model are you running on?"
    ```

    The response should come from your local model rather than from Anthropic's cloud. If you see an authentication error, double-check that `ANTHROPIC_BASE_URL` is set correctly and does not include a trailing slash.
  </Step>
</Steps>