open-agent-sdk

Open Agent SDK

Skill by ara.so — Daily 2026 Skills collection.

Open Agent SDK (@shipany/open-agent-sdk) is a fully open-source, in-process AI agent framework for TypeScript/Node.js. It runs the complete Claude Code agent engine directly — no local CLI subprocess required — making it suitable for cloud servers, serverless functions, Docker containers, and CI/CD pipelines. It is API-compatible with @anthropic-ai/claude-agent-sdk.

---

Installation

npm install @shipany/open-agent-sdk

Requires Node.js 18+.

---

Authentication & Configuration

Set the Anthropic API key as an environment variable:

export ANTHROPIC_API_KEY=your-api-key

Or use a third-party provider (e.g. OpenRouter):

export ANTHROPIC_BASE_URL=https://openrouter.ai/api
export ANTHROPIC_API_KEY=your-openrouter-key
export ANTHROPIC_MODEL=anthropic/claude-sonnet-4-6

These can also be passed programmatically via options.env or apiKey/baseURL in createAgent().

---

Core API

query({ prompt, options }) — Streaming, compatible with official SDK

Returns an AsyncGenerator<SDKMessage>. Drop-in replacement for @anthropic-ai/claude-agent-sdk.

import { query } from '@shipany/open-agent-sdk'

for await (const message of query({
  prompt: 'Find and fix the bug in auth.ts',
  options: {
    allowedTools: ['Read', 'Edit', 'Bash'],
    permissionMode: 'acceptEdits',
  },
})) {
  if (message.type === 'assistant' && message.message?.content) {
    for (const block of message.message.content) {
      if ('text' in block) process.stdout.write(block.text)
      else if ('name' in block) console.log(`\n[Tool used: ${block.name}]`)
    }
  } else if (message.type === 'result') {
    console.log(`\nDone: ${message.subtype}`)
  }
}

---

createAgent(options) — Reusable agent with session state

import { createAgent } from '@shipany/open-agent-sdk'

const agent = createAgent({
  model: 'claude-sonnet-4-6',
  systemPrompt: 'You are a senior TypeScript engineer. Be concise.',
  maxTurns: 20,
})

// Blocking call
const result = await agent.prompt('Read package.json and describe the project')
console.log(result.text)
console.log(`Tokens used: ${result.usage.input_tokens + result.usage.output_tokens}`)

// Streaming call
for await (const msg of agent.query('Now add JSDoc to all exported functions')) {
  if (msg.type === 'assistant' && msg.message?.content) {
    for (const block of msg.message.content) {
      if ('text' in block) process.stdout.write(block.text)
    }
  }
}

// Session management
const history = agent.getMessages()  // full conversation history
agent.clear()                        // reset session

---

Options Reference

Option	Type	Default	Description
model	string	claude-sonnet-4-6	Claude model ID
apiKey	string	ANTHROPIC_API_KEY env	API key
baseURL	string	Anthropic API	Override for third-party providers
cwd	string	process.cwd()	Working directory for file/shell tools
systemPrompt	string	—	Custom system prompt prepended to agent
tools	Tool[]	All built-in	Override the full tool list
allowedTools	string[]	all	Whitelist specific tools by name
permissionMode	string	bypassPermissions	acceptEdits, bypassPermissions, plan, default
maxTurns	number	100	Maximum agentic loop iterations
maxBudgetUsd	number	—	Spend cap in USD
mcpServers	object	—	MCP server configs (stdio/SSE/HTTP)
agents	object	—	Named subagent definitions
hooks	object	—	Lifecycle hooks: PreToolUse, PostToolUse, Stop
thinking	object	—	Extended thinking config
env	object	—	Environment variables passed to tools
resume	string	—	Resume prior session by session ID
canUseTool	function	—	Custom permission callback (tool, input) => boolean
includePartialMessages	boolean	false	Emit raw streaming events

---

Common Patterns

Multi-turn conversation with context

import { createAgent } from '@shipany/open-agent-sdk'

const agent = createAgent({ model: 'claude-sonnet-4-6' })

const r1 = await agent.prompt('Read src/index.ts and explain the architecture')
console.log(r1.text)

// Context from r1 is preserved automatically
const r2 = await agent.prompt('Refactor the error handling to use a Result type')
console.log(r2.text)

Restrict to read-only tools

import { query } from '@shipany/open-agent-sdk'

for await (const message of query({
  prompt: 'Review this codebase for security issues',
  options: {
    allowedTools: ['Read', 'Glob', 'Grep'],
    // No Write, Edit, or Bash — agent cannot modify files
  },
})) {
  if (message.type === 'result') console.log('Review complete')
}

Custom tools

import { createAgent, getAllBaseTools } from '@shipany/open-agent-sdk'

const dbQueryTool = {
  name: 'QueryDatabase',
  description: 'Run a read-only SQL query and return results as JSON',
  inputJSONSchema: {
    type: 'object',
    properties: {
      sql: { type: 'string', description: 'The SQL query to run' },
    },
    required: ['sql'],
  },
  get inputSchema() {
    return { safeParse: (v: unknown) => ({ success: true, data: v }) }
  },
  async prompt() { return this.description },
  async call(input: { sql: string }) {
    // Replace with your actual DB client
    const rows = [{ id: 1, name: 'Example' }]
    return { data: JSON.stringify(rows) }
  },
  userFacingName: () => 'QueryDatabase',
  isReadOnly: () => true,
  isConcurrencySafe: () => true,
  mapToolResultToToolResultBlockParam: (data: string, id: string) => ({
    type: 'tool_result' as const,
    tool_use_id: id,
    content: data,
  }),
}

const agent = createAgent({
  tools: [...getAllBaseTools(), dbQueryTool],
})

const result = await agent.prompt('How many users signed up in the last 7 days?')
console.log(result.text)

MCP server integration

import { createAgent } from '@shipany/open-agent-sdk'

const agent = createAgent({
  mcpServers: {
    filesystem: {
      command: 'npx',
      args: ['-y', '@modelcontextprotocol/server-filesystem', '/tmp'],
    },
    playwright: {
      command: 'npx',
      args: ['@playwright/mcp@latest'],
    },
  },
})

const result = await agent.prompt('List all .json files in /tmp')
console.log(result.text)

Subagents for parallel / delegated work

import { query } from '@shipany/open-agent-sdk'

for await (const message of query({
  prompt: 'Use the security-auditor agent to audit src/ for vulnerabilities',
  options: {
    allowedTools: ['Read', 'Glob', 'Grep', 'Agent'],
    agents: {
      'security-auditor': {
        description: 'Expert security auditor for TypeScript codebases.',
        prompt: 'Identify OWASP Top 10 vulnerabilities and suggest fixes.',
        tools: ['Read', 'Glob', 'Grep'],
      },
    },
  },
})) {
  if (message.type === 'assistant' && message.message?.content) {
    for (const block of message.message.content) {
      if ('text' in block) console.log(block.text)
    }
  }
}

Custom permission callback

import { createAgent } from '@shipany/open-agent-sdk'

const agent = createAgent({
  canUseTool: (toolName: string, input: unknown) => {
    // Prevent deletion commands
    if (toolName === 'Bash') {
      const cmd = (input as { command?: string }).command ?? ''
      if (cmd.includes('rm ') || cmd.includes('drop table')) return false
    }
    return true
  },
})

Lifecycle hooks

import { createAgent } from '@shipany/open-agent-sdk'

const agent = createAgent({
  hooks: {
    PreToolUse: async ({ tool, input }) => {
      console.log(`About to run tool: ${tool} with input:`, input)
    },
    PostToolUse: async ({ tool, output }) => {
      console.log(`Tool ${tool} finished`)
    },
    Stop: async ({ result }) => {
      console.log('Agent stopped. Final result:', result)
    },
  },
})

Resume a previous session

import { createAgent } from '@shipany/open-agent-sdk'

// First session
const agent1 = createAgent({ model: 'claude-sonnet-4-6' })
const r1 = await agent1.prompt('Read ARCHITECTURE.md')
const sessionId = r1.sessionId  // save this

// Later — resume where you left off
const agent2 = createAgent({
  model: 'claude-sonnet-4-6',
  resume: sessionId,
})
const r2 = await agent2.prompt('Now implement the TODO in section 3')

---

Built-in Tools Reference

Tool	Read-only	Description
Read	✅	Read files, images, PDFs with line numbers
Glob	✅	Find files by glob pattern
Grep	✅	Search file contents with regex (uses ripgrep)
WebFetch	✅	Fetch and parse web pages
WebSearch	✅	Web search
Write	❌	Create or overwrite files
Edit	❌	Precise string replacement in files
Bash	❌	Execute shell commands
Agent	—	Spawn subagents
TodoWrite	❌	Manage todo lists
NotebookEdit	❌	Edit Jupyter notebooks
TaskCreate/Update/List	—	Task management
TeamCreate/Delete	—	Agent team management
EnterPlanMode/ExitPlanMode	—	Plan approval workflow
EnterWorktree/ExitWorktree	—	Git worktree isolation
ListMcpResources/ReadMcpResource	✅	MCP resource access

---

Architecture: How It Differs from Official SDK

Official @anthropic-ai/claude-agent-sdk:

Your code → SDK → spawn cli.js subprocess → stdin/stdout JSON → Anthropic API

Open Agent SDK:

Your code → SDK → QueryEngine (in-process) → Anthropic API (direct HTTP)

This means:

• No CLI installation required in the deployment environment
• Works in serverless (AWS Lambda, Vercel, Cloudflare Workers with Node.js compat)
• Works in Docker with just npm install
• Works in CI/CD without CLI setup steps
• Programmatic access to the full agent engine

---

Troubleshooting

Error: ANTHROPIC_API_KEY is not set → Export the env var or pass apiKey directly in createAgent({ apiKey: process.env.MY_KEY }).

Agent exceeds maxTurns without completing → Increase maxTurns or narrow the task. Check message.subtype === 'max_turns' in the result.

Tool not found / allowedTools not working → Tool names are case-sensitive: 'Read', 'Edit', 'Bash', 'Glob', 'Grep', 'WebFetch', etc.

Using with OpenRouter or other providers → Set ANTHROPIC_BASE_URL to the provider's base URL and use their model string format, e.g. anthropic/claude-sonnet-4-6 for OpenRouter.

Agent modifies files unexpectedly → Use allowedTools: ['Read', 'Glob', 'Grep'] to restrict to read-only tools, or set permissionMode: 'plan' to require approval before edits.

MCP server fails to start → Ensure the MCP server package is installed or accessible via npx. Check command and args match what the MCP package expects.

TypeScript types missing → The package ships its own types. Ensure "moduleResolution": "bundler" or "node16" in tsconfig.json and "esModuleInterop": true.

---

Quick Reference

// Minimal one-shot agent
import { createAgent } from '@shipany/open-agent-sdk'
const agent = createAgent({ model: 'claude-sonnet-4-6' })
const { text } = await agent.prompt('Summarize README.md in 3 bullet points')
console.log(text)