Anthropic SDK Pro -- Claude AI for Next.js 16

When to Use

Trigger on any mention of: anthropic, claude, claude api, ai sdk, ai enhancement, streaming response, tool use, function calling, prompt engineering, token counting, ai rate limit, claude sonnet, claude haiku.

Reference Files

Load the relevant file(s) from references/ based on the task:

File	Content
`setup.md`	SDK compatibility (version matrix, runtime), client init, env vars, API key security
`messages-api.md`	Single/multi-turn requests, system prompts, parameters, response structure, models + pricing
`streaming.md`	`stream: true` vs `.stream()`, SSE route handlers, client-side consumption, event types
`tool-use.md`	JSON Schema tools, agentic loop, Zod-based tools, tool choice, structured output via tools
`error-handling.md`	SDK error classes, auto retry behavior, timeout config, request ID tracking, Route Handler pattern
`token-management.md`	Token estimation, countTokens() API, usage tracking, max_tokens strategies, optimization tips
`cv-ai-patterns.md`	6 prompts: bullet enhance, summary, skills, translate, ATS, cover letter + endpoint table
`errors.md`	18 error entries (AI-001 through AI-018) with exact messages, causes, and fixes
`rate-limiting.md`	Tier tables (1-4), RPM/ITPM/OTPM limits, cache-aware ITPM, app-level usage tracking
`cost-optimization.md`	Model selection, prompt caching (90% savings), Batch API (50% off), CViet cost estimates
`vercel-ai-sdk.md`	@ai-sdk/anthropic, streamText, generateText, useChat, structured output, extended thinking
`security.md`	Input sanitization, prompt injection prevention, output validation, defense-in-depth
`prompt-engineering.md`	General principles, system prompt template, temperature guide, output format, structured output

Error Quick Lookup

ID	Error	Fix
AI-001	Error objects crossing RSC boundary	Catch + return `{ error: e.message }` plain object
AI-002	401 authentication_error	Check `ANTHROPIC_API_KEY` env var
AI-003	429 rate_limit_error	Implement exponential backoff, check tier limits
AI-004	529 overloaded_error	Retry with delay, SDK auto-retries 2x
AI-005	400 invalid_request_error	Validate all params before sending
AI-006	Context length exceeded	Truncate input, use `countTokens()` before sending
AI-007	Streaming timeout	Set `maxDuration` in route config, use `stream: true`
AI-008	JSON parsing failure	Use regex extraction or structured outputs
AI-009	Model not found	Use exact model ID: `claude-sonnet-4-6`
AI-010	Content policy violation	Review input, handle gracefully
AI-011	Empty content response	Check `stop_reason`, increase `max_tokens`
AI-012	TypeScript compilation errors	Pin SDK version, check changelog
AI-013	Network connection error	Check connectivity, SDK auto-retries 2x
AI-014	Non-streaming request too long	Use `stream: true` or increase timeout
AI-015	Invalid API key format	Must start with `sk-ant-api03-`, regenerate key
AI-016	Concurrent request limit	Queue requests, upgrade API tier
AI-017	Output truncated	Increase `max_tokens`, check `stop_reason`
AI-018	Prompt caching minimum size	Content must exceed 1024-4096 token threshold

Key Patterns

CV Bullet Enhancement

// app/api/ai/enhance/route.ts
const response = await anthropic.messages.create({
  model: 'claude-sonnet-4-6',
  max_tokens: 1024,
  system: 'You are an expert CV writer. Rewrite bullet points to be impactful and ATS-friendly.',
  messages: [{ role: 'user', content: `Rewrite these bullets:\n${bullets.join('\n')}` }]
})

Streaming in Route Handler

export async function POST(req: Request) {
  const stream = await anthropic.messages.stream({
    model: 'claude-sonnet-4-6',
    max_tokens: 1024,
    messages: [{ role: 'user', content: prompt }]
  })
  return new Response(stream.toReadableStream())
}

Usage Tracking (Free/Pro Gating)

if (user.plan === 'FREE' && user.aiUsageThisMonth >= 3) {
  return NextResponse.json({ error: 'AI limit reached' }, { status: 429 })
}
await db.user.update({ where: { id: user.id }, data: { aiUsageThisMonth: { increment: 1 } } })

anthropic-sdk-pro