Toolofficial publisher

Tools Claude Agent SDK

Tool : A precise action the AI can call · requires anthropic

Claude Code tools exposed in your own agent, with the permissions system.

The shortest path to a serious agent with native permissions.

Docs

Overall score

6.9 /10

Security: one caveat, see the card

Utility

8.0

Reliability

6.0

Security

One caveat

Ease

5 min · Technical

Security reflects where data goes, requested permissions and incidents on record at the verification date. It is not a guarantee of zero risk: check the deployment mode and the connections you enable.

What it does, concretely

Building an AI agent capable of reading files, running commands, or editing code usually requires implementing the planning and tool-calling loop yourself. Anthropic's Agent SDK solves this by providing the agent loop, context management, and built-in tools that power Claude Code, packaged as a Python or TypeScript library.

In practice, the SDK exposes native tools for reading, writing, and editing files, running commands, searching the web, and connecting to external sources via the MCP protocol. It also allows spawning specialized subagents for focused subtasks, running custom code at key points in the agent lifecycle through hooks, and finely controlling which tools run automatically or require human approval through the permissions system.

This SDK is aimed at Python or TypeScript developers who want to build production agents without starting from scratch. It is particularly suited to those who want the same capabilities as Claude Code within their own application, with full control over permissions and agent behavior.

It is not suited to those looking for an interactive terminal interface (that is the role of the Claude Code CLI), nor to those who want to call the Anthropic API directly while managing the tool loop themselves (that is the role of the Client SDK). For asynchronous or long-running agents without infrastructure management, Managed Agents are a separate alternative.

What it offers

·Built-in agent loop: the SDK handles planning, tool calling, and task completion decisions without you having to implement them.
·Native tools: file reading, writing, editing, command execution, and web search are available out of the box.
·Permissions system: control which tools run automatically and which require human approval.
·Subagents: spawn specialized agents for focused subtasks within a single workflow.
·Hooks: run custom code at key points in the agent lifecycle.
·MCP connectivity: integrate external tools and data sources via the Model Context Protocol.
·Sessions: maintain context across exchanges, resume or fork an existing session.
·Plugins: bundle skills, agents, hooks, and MCP servers into reusable packages loaded by local path.

Our tips

If you are developing in a language other than Python or TypeScript, you can drive the same agent loop by running the CLI as a subprocess with the -p flag and --output-format json.
Skills, commands, and memory placed in your project's .claude/ folder or in ~/.claude/ are loaded automatically by the SDK, just as in Claude Code.
Use hooks to insert business logic or security checks at specific points in the agent lifecycle without modifying the main loop.
The permissions system lets you define which tools require human approval before execution: configure it carefully based on the sensitivity of the operations your agent performs.
For complex workflows, subagents allow you to delegate specialized subtasks and orchestrate multiple agents in parallel, as the Claude Code team does with its agent harness design.

Watch out

Authentication via claude.ai is not permitted for third-party developers: you must use an API key, which means usage costs to monitor depending on request volume.
The SDK is only available in Python and TypeScript; integrating it in another language requires a workaround via the CLI as a subprocess, which adds complexity.

In practice

Price: Free
Commercial use: Commercial use: yes
Available on: sdk-python, sdk-ts
Host: anthropic
Licence: MIT
FR interface: No

Security and data

Rating: Security: one caveat, see the card
Capabilities: readswritesexecutesnetwork
Your data: Data stays local
Maintenance: Active

Links and repository checked on 25 October 2026 · added on 1 September 2025

How to install Tools Claude Agent SDK

1. Make sure you have a compatible Python or TypeScript environment. 2. Install the SDK library for your language (Python or TypeScript). 3. Obtain an Anthropic API key and set it in your environment (never commit it to your code repository). 4. Follow the official Quickstart to build your first bug-finding and bug-fixing agent. 5. Configure the permissions system from the start to control which tools run automatically.

pip install claude-agent-sdk

Official documentation ↗

In the same category

Whole category Orchestrate →