Claude Code AI Provider

A Vercel AI SDK community provider for Claude Code integration using the official TypeScript SDK.

This provider allows you to use Claude Code as a language model within the Vercel AI SDK ecosystem, powered by the official @anthropic-ai/claude-code TypeScript SDK for reliability and maintainability.

Features

• 🔌 Vercel AI SDK Compatible: Drop-in replacement for other AI providers
• 🚀 Official SDK Integration: Built on @anthropic-ai/claude-code TypeScript SDK
• 📋 Structured Output: Generate type-safe JSON objects using schemas
• 📺 Streaming Support: Real-time streaming responses
• 🛠️ Tool Support: Access to Claude Code's powerful development tools
• ⚙️ Configurable: Extensive configuration options
• 🔄 Session Management: Support for multi-turn conversations

Installation

npm install claude-code-ai-provider @anthropic-ai/claude-code
# or
pnpm add claude-code-ai-provider @anthropic-ai/claude-code
# or
yarn add claude-code-ai-provider @anthropic-ai/claude-code

Note: @anthropic-ai/claude-code is a peer dependency. This ensures you can always use the latest version of Claude Code SDK without waiting for this provider to update.

Prerequisites

• Claude Code CLI must be installed and available in your PATH
• Valid Anthropic API key configured for Claude Code
• Node.js 18+

Install Claude Code CLI:

npm install -g @anthropic-ai/claude-code

Configure your API key:

export ANTHROPIC_API_KEY=your-api-key

Quick Start

Basic Usage

import { createClaudeCode } from 'claude-code-ai-provider'
import { generateText } from 'ai'

const claudeCode = createClaudeCode()

const { text } = await generateText({
  model: claudeCode('sonnet'), // or 'opus', 'haiku', or specific version like 'claude-sonnet-4-20250514'
  prompt: 'Write a function to calculate fibonacci numbers',
})

console.log(text)

Streaming

import { createClaudeCode } from 'claude-code-ai-provider'
import { streamText } from 'ai'

const claudeCode = createClaudeCode()

const { textStream } = await streamText({
  model: claudeCode('sonnet'),
  prompt: 'Create a React component for a todo list',
})

for await (const delta of textStream) {
  process.stdout.write(delta)
}

Structured Output (Object Generation)

import { generateObject } from 'ai'
import { z } from 'zod'
import { createClaudeCode } from 'claude-code-ai-provider'

const claudeCode = createClaudeCode()

const result = await generateObject({
  model: claudeCode('sonnet'),
  schema: z.object({
    recipe: z.object({
      name: z.string(),
      ingredients: z.array(
        z.object({
          name: z.string(),
          amount: z.string(),
        }),
      ),
      steps: z.array(z.string()),
      prepTime: z.string(),
      cookTime: z.string(),
      servings: z.number(),
    }),
  }),
  prompt: 'Generate a recipe for chocolate chip cookies',
})

console.log(JSON.stringify(result.object, null, 2))

Streaming Object Generation

import { streamObject } from 'ai'
import { z } from 'zod'
import { createClaudeCode } from 'claude-code-ai-provider'

const claudeCode = createClaudeCode()

const { partialObjectStream } = await streamObject({
  model: claudeCode('sonnet'),
  schema: z.object({
    story: z.object({
      title: z.string(),
      characters: z.array(
        z.object({
          name: z.string(),
          role: z.string(),
        }),
      ),
    }),
  }),
  prompt: 'Create a fantasy story outline',
})

for await (const partialObject of partialObjectStream) {
  console.log(JSON.stringify(partialObject, null, 2))
}

Provider Configuration

import { createClaudeCode } from 'claude-code-ai-provider'

const claudeCode = createClaudeCode({
  // Path to Claude Code executable (default: 'claude')
  claudeCodeExecutable: '/usr/local/bin/claude',

  // Working directory for Claude Code execution
  cwd: '/path/to/project',

  // Environment variables
  env: {
    ANTHROPIC_API_KEY: 'your-api-key',
  },

  // Timeout in milliseconds (default: 5 minutes)
  timeout: 300000,

  // Enable verbose logging
  verbose: true,

  // Permission mode
  permissionMode: 'acceptEdits',

  // Allowed tools
  allowedTools: ['Read', 'Write', 'Bash'],

  // MCP configuration
  mcpConfig: '/path/to/mcp-config.json',
})

Model-Specific Settings

const model = claudeCode('sonnet', {
  // Maximum conversation turns
  maxTurns: 5,

  // Custom system prompt
  systemPrompt:
    'You are a senior software engineer specializing in TypeScript.',

  // Session management
  sessionId: 'my-session-id',
  continue: true,

  // Tool restrictions
  allowedTools: ['Read', 'Write'],
  disallowedTools: ['Bash'],
})

Configuration Options

Provider Settings

| Option | Type | Default | Description | | ---------------------- | ------------------------ | --------------- | ---------------------------------- | | claudeCodeExecutable | string | "claude" | Path to Claude Code executable | | cwd | string | process.cwd() | Working directory | | env | Record<string, string> | process.env | Environment variables | | timeout | number | 300000 | Timeout in milliseconds | | verbose | boolean | false | Enable verbose logging | | permissionMode | string | "default" | Permission mode for tool execution | | mcpConfig | string | - | MCP configuration file path | | allowedTools | string[] | - | Default allowed tools | | disallowedTools | string[] | - | Default disallowed tools |

Model Settings

| Option | Type | Default | Description | | ---------------------- | ---------- | ------- | ---------------------------------------- | | maxTurns | number | - | Maximum conversation turns | | systemPrompt | string | - | Custom system prompt | | appendSystemPrompt | string | - | Additional system prompt | | sessionId | string | - | Session ID for conversation continuation | | continue | boolean | false | Continue latest conversation | | allowedTools | string[] | - | Allowed tools for this model | | disallowedTools | string[] | - | Disallowed tools for this model | | permissionPromptTool | string | - | MCP tool for permission prompts |

Supported Models

Model Aliases (Latest Versions)

• sonnet - Claude Sonnet 4 (latest: claude-sonnet-4-20250514)
• opus - Claude Opus 4 (latest: claude-opus-4-20250514)
• haiku - Claude 3.5 Haiku (latest: claude-3-5-haiku-20241022)

Specific Model Versions

• claude-sonnet-4-20250514 - Claude Sonnet 4 (May 2025)
• claude-opus-4-20250514 - Claude Opus 4 (May 2025)
• claude-3-5-sonnet-20241022 - Claude 3.5 Sonnet (October 2024)
• claude-3-5-haiku-20241022 - Claude 3.5 Haiku (October 2024)

Custom Models

Any string can be used as a model ID for custom or future models.

Advanced Usage

Multi-turn Conversations

import { createClaudeCode } from 'claude-code-ai-provider'
import { generateText } from 'ai'

const claudeCode = createClaudeCode()

// First turn
const { text: response1 } = await generateText({
  model: claudeCode('sonnet', {
    sessionId: 'my-project-session',
  }),
  prompt: 'Create a new React project structure',
})

// Continue the same session
const { text: response2 } = await generateText({
  model: claudeCode('sonnet', {
    sessionId: 'my-project-session',
    continue: true,
  }),
  prompt: 'Now add TypeScript configuration',
})

Tool Management

const model = claudeCode('sonnet', {
  // Allow specific tools
  allowedTools: [
    'Read',
    'Write',
    'Bash(npm install)', // Allow specific bash commands
    'mcp__filesystem__read_file', // Allow MCP tools
  ],

  // Disallow specific tools
  disallowedTools: [
    'Bash(rm -rf)', // Disallow dangerous commands
  ],

  // Permission mode
  permissionMode: 'acceptEdits',
})

MCP Integration

const claudeCode = createClaudeCode({
  mcpConfig: './mcp-servers.json',
  allowedTools: ['mcp__github__create_issue', 'mcp__slack__send_message'],
  permissionPromptTool: 'mcp__permissions__approve',
})

Examples

The examples/ directory contains complete TypeScript examples:

Text Generation

• generate-text.ts - Basic text generation with configuration
• stream-text.ts - Real-time streaming responses

Structured Output

• generate-object.ts - Simple structured output (recipe)
• generate-object-complex.ts - Complex nested schemas (project config)
• stream-object.ts - Streaming object generation

Run examples:

cd examples
pnpm install

# Run TypeScript examples directly
pnpm generate "Write a hello world function"
pnpm stream "Explain async/await"
pnpm generate-object

Error Handling

import { createClaudeCode, isClaudeCodeError } from 'claude-code-ai-provider'
import { generateText } from 'ai'

const claudeCode = createClaudeCode()

try {
  const { text } = await generateText({
    model: claudeCode('sonnet'),
    prompt: 'Write some code',
  })
  console.log(text)
} catch (error) {
  if (isClaudeCodeError(error)) {
    console.error(`Claude Code Error (${error.type}): ${error.message}`)
    if (error.sessionId) {
      console.error(`Session ID: ${error.sessionId}`)
    }
  } else {
    console.error('Unexpected error:', error)
  }
}

Error Types

• timeout_error - Request exceeded timeout limit
• process_error - Claude Code execution failed
• parsing_error - Failed to parse response
• invalid_response - Invalid response format

Development

Building

pnpm build

Testing

pnpm test

Linting & Formatting

pnpm lint       # Run Biome linter
pnpm format     # Check formatting
pnpm format:fix # Fix formatting

Architecture

This provider is built on top of the official @anthropic-ai/claude-code TypeScript SDK:

claude-code-ai-provider
├── ClaudeCodeLanguageModel (Vercel AI SDK interface)
├── ClaudeCodeSDK (Official SDK wrapper)
├── MessageConverter (SDK ↔ AI SDK conversion)
└── StreamConverter (Streaming support)

Contributing

Contributions are welcome! Please read the contributing guidelines and submit pull requests.

License

MIT License - see LICENSE file for details.

Related

• Vercel AI SDK - The AI SDK this provider integrates with
• Claude Code - The Claude Code CLI tool
• @anthropic-ai/claude-code - Official TypeScript SDK
• Anthropic - The company behind Claude