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

# Use Claude Code with Parallel AI

> Parallel AI exposes an Anthropic Messages API-compatible endpoint, so Claude Code and any Anthropic SDK can use Parallel AI as their LLM provider.

## Overview

Parallel AI's `/api/v0/claude` endpoints are compatible with the Anthropic Messages API. Point **Claude Code** — or any Anthropic SDK — at Parallel AI and it works out of the box, billed against your Parallel AI credits like any other API usage.

| Endpoint                                       | Description                                                        |
| ---------------------------------------------- | ------------------------------------------------------------------ |
| `POST /api/v0/claude/v1/messages`              | Create a message (streaming and non-streaming, tool use supported) |
| `POST /api/v0/claude/v1/messages/count_tokens` | Estimate token count for a request                                 |
| `GET /api/v0/claude/v1/models`                 | List available models                                              |

<Tip>
  The `model` field accepts **any** Parallel AI model — not just Anthropic
  models — so you can run Claude Code on any model the platform offers.
</Tip>

## Authentication

Authentication works exactly like the rest of the API: send your API key as `Authorization: Bearer <key>` or `x-api-key: <key>`.

Personal Access Keys (`pak_...`) also work but require the `X-Company-ID` header, which Claude Code cannot send by default — use a **Company API Key** for Claude Code.

## Setting up Claude Code

<Steps>
  <Step title="Install Claude Code">
    Skip this if it's already installed.

    ```bash theme={null}
    # macOS / Linux / WSL — native installer
    curl -fsSL https://claude.ai/install.sh | bash

    # Or via npm (requires Node.js 18+)
    npm install -g @anthropic-ai/claude-code
    ```

    On Windows (PowerShell):

    ```powershell theme={null}
    irm https://claude.ai/install.ps1 | iex
    ```

    Verify the install with `claude --version`. See the [official docs](https://code.claude.com/docs) for more install options.
  </Step>

  <Step title="Get a Company API Key">
    In the dashboard, go to **Integrations** → **API Keys** → **Generate New Key**.
  </Step>

  <Step title="Configure Claude Code">
    Either export environment variables in your shell:

    ```bash theme={null}
    export ANTHROPIC_BASE_URL="https://api.parallellabs.app/api/v0/claude"
    export ANTHROPIC_AUTH_TOKEN="YOUR_COMPANY_API_KEY"
    claude
    ```

    Or persist the settings in `~/.claude/settings.json` (applies to all projects) or `.claude/settings.local.json` in a project. Set the model variables to any Parallel AI model names — for example:

    ```json theme={null}
    {
      "env": {
        "ANTHROPIC_BASE_URL": "https://api.parallellabs.app/api/v0/claude",
        "ANTHROPIC_AUTH_TOKEN": "YOUR_COMPANY_API_KEY",
        "ANTHROPIC_MODEL": "qwen3.8-max",
        "ANTHROPIC_DEFAULT_OPUS_MODEL": "kimi-k3",
        "ANTHROPIC_DEFAULT_SONNET_MODEL": "qwen3.8-max",
        "ANTHROPIC_DEFAULT_HAIKU_MODEL": "deepseek-v4-flash-0731"
      }
    }
    ```
  </Step>

  <Step title="Verify">
    Run `/status` inside Claude Code — it should show your custom base URL — or test the endpoint directly:

    ```bash theme={null}
    curl -X POST "https://api.parallellabs.app/api/v0/claude/v1/messages" \
      -H "Authorization: Bearer YOUR_COMPANY_API_KEY" \
      -H "anthropic-version: 2023-06-01" \
      -H "Content-Type: application/json" \
      -d '{"model": "qwen3.8-max", "max_tokens": 64, "messages": [{"role": "user", "content": "Hello"}]}'
    ```
  </Step>
</Steps>

## Model names

The `model` field accepts any model name from `GET /api/v0/models` — any provider, not just Anthropic. When Claude Code sends its default Anthropic model IDs (e.g. `claude-sonnet-4-5-20250929`), they are routed by tier to Parallel AI models:

* **Opus-class** requests run on the platform's highest-capability model
* **Sonnet-class** requests run on the balanced default model
* **Haiku-class** requests (background tasks) run on the fastest low-cost model

Set `ANTHROPIC_MODEL` and the `ANTHROPIC_DEFAULT_*_MODEL` variables to exact Parallel AI model names for full control over routing.

Optionally set `CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY=1` so Claude Code lists Parallel AI's models in its `/model` picker.

## Using the Anthropic SDK directly

<CodeGroup>
  ```python Python theme={null}
  import anthropic

  client = anthropic.Anthropic(
      base_url="https://api.parallellabs.app/api/v0/claude",
      auth_token="YOUR_COMPANY_API_KEY",
  )

  response = client.messages.create(
      model="kimi-k3",
      max_tokens=1024,
      messages=[{"role": "user", "content": "Hello"}],
  )
  print(response.content[0].text)
  ```

  ```typescript TypeScript theme={null}
  import Anthropic from "@anthropic-ai/sdk";

  const client = new Anthropic({
    baseURL: "https://api.parallellabs.app/api/v0/claude",
    authToken: "YOUR_COMPANY_API_KEY",
  });

  const response = await client.messages.create({
    model: "kimi-k3",
    max_tokens: 1024,
    messages: [{ role: "user", content: "Hello" }],
  });
  console.log(response.content[0].text);
  ```
</CodeGroup>

## Notes and limitations

* **Streaming** is fully supported (Anthropic SSE event format with keep-alive pings).
* **Tool use** is supported: Claude Code's client-side tools (file edits, bash, etc.) are forwarded to the model and executed locally by Claude Code — never on Parallel AI's servers.
* Server-side Anthropic tools (web search, code execution), extended thinking blocks, and prompt caching directives are accepted but ignored.
* Usage is billed per request based on real token counts and appears in your usage dashboard under the chat feature.