↓ Skip to main content
9 – 17 UHR +49 8031 3508270 LUITPOLDSTR. 9, 83022 ROSENHEIM
DE / EN
Coding agent Anthropic United States Terminal, IDE, Desktop, Web

Claude Code

Claude Code is Anthropic's agentic coding harness for terminal, IDE, desktop and web, with tools, context, permissions and subagents for developers.

Website Docs GitHub Changelog
Licence Proprietary
First release February 2025 (repository created)
Interfaces Terminal (CLI), VS Code, JetBrains, Desktop app, Web
Platforms macOS, Linux, Windows
Instruction file CLAUDE.md

The seven building blocks in Claude Code

How Claude Code implements the building blocks of an AI harness

  1. 1

    Agent loop

    Claude Code runs an agentic loop of gathering context, taking action and verifying results until a task is complete, a step limit is hit, or it needs to ask a question.

  2. 2

    Tools and execution environment

    Built-in tools cover file operations, search, command execution and web access; MCP lets you connect any external system as an additional tool.

  3. 3

    Context management

    Project files, git state, CLAUDE.md and tool results fill the context window; Claude Code compacts and clears old output automatically as it fills up.

  4. 4

    State and memory

    CLAUDE.md provides persistent project instructions, and auto memory saves learnings automatically to MEMORY.md across sessions.

  5. 5

    Permissions

    A tiered permission system with modes such as Manual, Auto, Plan and Accept Edits, plus fine-grained allow/ask/deny rules, decides what runs without asking.

  6. 6

    Verification and feedback

    Checkpoints snapshot file changes before every prompt, so changes can be undone with /rewind; running tests and builds provides content verification.

  7. 7

    Orchestration

    Subagents with their own context window, tools and permissions handle subtasks in parallel and return only a summary to the main conversation.

Models and model access

Which models the harness uses and how they are obtained

Models and model access
Models Claude models (Opus, Sonnet, Haiku)
Access routes Claude API, Amazon Bedrock, Google Cloud's Agent Platform (formerly Vertex AI), Microsoft Foundry
Free choice of model No
Custom endpoint / LLM gateway Yes
Claude Code can be pointed at a self-hosted LLM gateway via the ANTHROPIC_BASE_URL environment variable, as long as the gateway exposes an API format Anthropic supports.
Features
MCP servers Sub-agents Hooks Skills Plugins / extensions Sandbox Headless / CI SDK IDE integration Checkpoints Resume sessions
Running it with herdr
Supported by herdr Yes
State detection Screen manifest
Integration Session
According to herdr's own documentation, it detects Claude Code sessions via a screen manifest and integrates them at the session level.
More about herdr →

Enterprise use

Data processing, controls and pricing model according to the vendor

Data processing and retention

For commercial access (Team, Enterprise, API), Anthropic's own documentation states it does not train models on code or prompts from Claude Code unless a customer has explicitly opted in. Standard retention is 30 days; Zero Data Retention is available separately for qualified Claude for Enterprise accounts.

Identity and SSO

Claude for Enterprise offers SSO, domain capture and role-based permissions according to the vendor; Amazon Bedrock, Google Cloud's Agent Platform and Microsoft Foundry each rely on their respective cloud provider's identity and IAM mechanisms.

Admin policies

According to the documentation, managed settings can be distributed to every developer machine as a managed-settings.json file via MDM or device management, and take precedence over every other configuration level.

Telemetry

Usage metrics and error reports can be disabled individually or together via environment variables such as DISABLE_TELEMETRY or CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC; per the documentation, they are off by default when accessed through a cloud provider.

Pricing model

Access runs through a subscription (Pro, Max, Team, Enterprise) or usage-based via an API key or the billing of the chosen cloud provider – see the vendor for current terms.

Strengths
  • Tight integration with terminal, git, IDEs and CI/CD pipelines
  • Broad extensibility via MCP, hooks, skills, plugins and subagents
  • Connects to three major cloud platforms (Bedrock, Google Cloud's Agent Platform, Microsoft Foundry) for regional model deployment
  • Fine-grained, version-controllable permission system with several automation modes
  • Checkpoints allow low-risk, iterative work with the ability to roll back
  • Available across many surfaces from terminal to web and Slack
Limitations
  • Proprietary software with no open licence; the source code is not publicly inspectable
  • Without a custom adapter it is built primarily around Claude models, with no cross-vendor model routing in the core product
  • According to the documentation, Zero Data Retention is not included in the standard Enterprise plan and requires separate enablement
  • According to the documentation, the native sandbox does not run on native Windows and requires WSL2 there

Use cases

Typical tasks for this harness

Debugging and bug fixes spanning multiple files
Feature development with an automated test-and-build cycle
Automated code review and pull-request creation in CI/CD pipelines
Non-interactive automation via claude -p in scripts and GitHub Actions
Refactoring and migration work in existing codebases
Parallel work on large tasks using multiple subagents

Claude Code is Anthropic’s agentic coding harness: it reads codebases, edits files, runs commands and connects development tools to turn the Claude language model into a working AI harness for developers. It is built by Anthropic and available in the terminal, in VS Code and JetBrains IDEs, as a desktop app, and in the browser. As of 20 September 2026.

What is Claude Code?

Claude Code comes from Anthropic and, according to its GitHub repository, was first published in February 2025. The software is proprietary: according to the licence file in the official repository, use is governed by Anthropic’s commercial terms of service, and the source code itself is not openly licensed. Claude Code is installed with curl -fsSL https://claude.ai/install.sh | bash (macOS, Linux, WSL) or via Homebrew, WinGet, or package managers such as apt and dnf, and started by running claude inside a project directory. Beyond the terminal CLI, there are extensions for VS Code and JetBrains, a standalone desktop app, and browser access at claude.ai/code.

Claude Code as an AI harness

According to its own documentation, Claude Code serves as an “agentic harness” around the Claude model: it provides the tools, context management and execution environment that turn the model into a coding agent. Its agentic loop cycles through gathering context, taking action and verifying results until a task is complete. Built-in tools cover file operations, code search, command execution and web access; the Model Context Protocol (MCP) lets you connect additional servers via claude mcp add. Project instructions live in a CLAUDE.md file at the project root, and Claude Code can also read an existing AGENTS.md. Permissions are enforced through a tiered system of permission modes (Manual, Auto, Plan, Accept Edits) plus fine-grained allow/ask/deny rules in .claude/settings.json. For complex tasks, Claude Code spins up subagents with their own context window, tools and permissions, which return only a summary to the main conversation.

Models and model access

Claude Code uses models from Anthropic’s Claude model family — Opus for complex architectural decisions, Sonnet for most coding tasks, Haiku for background processes. Access runs through the Claude API, Amazon Bedrock, Google Cloud’s Agent Platform (formerly Vertex AI), or Microsoft Foundry, each with its own environment variables for authentication and model pinning. For organisations running their own LLM gateway, Anthropic documents the ANTHROPIC_BASE_URL environment variable, which points Claude Code at a self-hosted gateway as long as it exposes one of the supported API formats. This is the route innFactory’s AI Gateway uses: Claude Code runs entirely through the gateway, which acts as a central proxy handling budgets, cost centres, guardrails and authentication via Microsoft Entra ID – with Claude obtained via AWS Bedrock or Google Cloud if preferred.

Enterprise use: data protection and governance

According to Anthropic’s data-usage documentation, for commercial access (Team, Enterprise, API, third-party platforms) the company does not train models on submitted code or prompts unless an organisation has explicitly opted in. Standard retention is 30 days; Zero Data Retention can be separately enabled by Anthropic for qualified Claude for Enterprise accounts and covers only Anthropic’s direct platform, not Bedrock, Google Cloud’s Agent Platform, or Microsoft Foundry. For enterprise governance, Anthropic documents managed settings that can be centrally distributed via MDM as a managed-settings.json file and take precedence over every other configuration level, along with SSO and role-based permissions under Claude for Enterprise. Telemetry can be disabled individually via environment variables such as DISABLE_TELEMETRY. Important in practice: source code and prompts are sent to the chosen model provider, which should be mapped in advance in an organisation’s records of processing activities.

Orchestrating Claude Code with herdr

According to the documentation of the terminal orchestrator herdr, it detects Claude Code sessions via a screen manifest and integrates them at the session level, without relying on Claude Code’s own lifecycle hooks. This makes it possible to monitor and control several Claude Code instances alongside other coding agents from one shared terminal interface.

Where it fits: developer harness and enterprise harness

Claude Code is a harness built for a very specific environment: one repository, one terminal, one developer. For business users outside development, CompanyGPT is the matching harness: tools via MCP servers and n8n workflows, context via companyRAG, roles from Entra ID or Keycloak, and audit logs for business departments. The AI Gateway forms the shared cost and governance layer for both worlds — developer harnesses like Claude Code as well as enterprise harnesses like CompanyGPT. For an overview of other coding agents and orchestrators, see our overview of all AI harnesses. We help organisations adopt Claude Code and comparable harnesses securely, with proper governance and centralised cost control.

From developer harness to enterprise harness

Coding harnesses are tools for developers. Business users, cost control and governance need two more layers.

CompanyGPT – the harness for business users

CompanyGPT gives any language model tools via MCP servers, integrations and n8n, company knowledge via companyRAG, roles from Entra ID or Keycloak, and audit logs – operated in your Azure environment or sovereignly on STACKIT.

Discover CompanyGPT
AI Gateway – cost and governance for every agent

The AI Gateway bundles the model access of your tools and coding agents behind an OpenAI-compatible API: authentication via Entra ID, budgets and cost centres per team, user and agent, guardrails and complete logging.

View the AI Gateway

Frequently Asked Questions

What is Claude Code?

Claude Code is Anthropic's agentic coding harness. It reads codebases, edits files, runs commands and connects development tools to turn the Claude language model into a working coding agent. It is available in the terminal, in VS Code and JetBrains IDEs, as a desktop app, and in the browser.

Is Claude Code open source?

No. Claude Code is proprietary software from Anthropic; according to the licence file in its official GitHub repository, use is governed by Anthropic's commercial terms of service. The repository serves issue tracking and documentation, not free reuse of the source code.

Which models does Claude Code support?

Claude Code uses Claude models from the Opus, Sonnet and Haiku families. Access runs through the Claude API or through Amazon Bedrock, Google Cloud's Agent Platform, or Microsoft Foundry, each with its own environment variables for configuration and model pinning.

Can Claude Code be used in a GDPR-compliant way?

According to the vendor's documentation, Anthropic does not train models on submitted code for commercial access, and Zero Data Retention is available for qualified Enterprise accounts. Prompts and source code are sent to the chosen model provider in the process, so organisations should map this processing in their records of processing activities in advance and verify current terms with Anthropic or the chosen cloud provider.

Does Claude Code work with herdr?

Yes. According to herdr's documentation, herdr detects Claude Code sessions via a screen manifest and integrates them at the session level when orchestrating multiple coding agents.

What is the difference between Claude Code and the Claude Agent SDK?

Claude Code is the finished application for developers. According to Anthropic, the Claude Agent SDK exposes the same agentic loop, tools and context management as building blocks for creating custom, programmable agents in Python or TypeScript.

More AI harnesses

Compare other harnesses

Coding agent OpenAI

Codex

Codex is OpenAI's agentic coding harness for the terminal, IDE and cloud. Features, sandbox modes, model …

Coding agent Earendil Inc.

Pi

Pi is the open-source, minimalist coding agent harness from Earendil Inc. for the terminal – extensible via …

Orchestrator Herdr, Inc.

herdr

herdr is the open-source terminal runtime by Herdr, Inc. for AI coding agents such as Claude Code or Codex, …

Introduce coding agents safely in your company

We support selection, model access in the EU, permission design and cost control.

Request consultation