Understand and use Anthropic’s Claude safely: current models, apps, API, Claude Code, MCP, skills, data protection, cost and limitations.
Roland Schellhorn
Claude is a family of large language models developed by the US company Anthropic. You can use Claude as an assistant in the browser and in apps, build it into your own applications through the API, or let it work directly in your repository with Claude Code. For companies, the real question is not whether Claude writes good text, but where the model fits sensibly into processes, data and software.
This guide explains the models, access options and tools, and shows what to watch for in data protection, cost and security. Information current as of September 2026.
According to Anthropic’s models overview, all current Claude models accept text and images as input, produce text output, are multilingual and can call tools (“tool use”). That makes Claude more than a chatbot: it is a foundation for assistants, automations and agents that work through multistep tasks on their own.
Anthropic offers several models that differ in capability, speed and price:
According to Anthropic, the three larger models handle up to 1 million tokens of context and Haiku 4.5 up to 200,000 tokens. That is enough to include lengthy contracts, manuals or large parts of a codebase in a single request. More context also means higher cost, though, and it does not replace choosing the relevant information carefully.
With claude.ai and the desktop and mobile apps, employees work with Claude directly: drafting text, analyzing documents, evaluating spreadsheets or creating presentations. Team and Enterprise plans add central administration and commercial terms.
The Claude API lets you build Claude into your own products and processes, such as a customer portal, a support workflow or a content pipeline. Besides Anthropic’s platform, the models are available through Amazon Bedrock, Google Cloud Vertex AI and Microsoft Foundry. This matters when data residency and billing should run through an existing cloud contract.
Claude Code is an agentic coding tool. It reads your codebase, edits files, runs commands and works with Git. It runs in the terminal, in VS Code and JetBrains, as a desktop app and in the browser. Project rules live in a CLAUDE.md file; Claude Code can also read an existing AGENTS.md. Our article on ChatGPT Codex describes how a comparable coding agent is used day to day. The principles on reviews and approvals there apply to Claude Code just as well.
The Model Context Protocol (MCP) is an open standard for connecting AI applications to data sources and tools, such as ticketing systems, databases or internal APIs. Anthropic originally introduced MCP; other vendors and development tools now support it as well.
Agent Skills package recurring knowledge as a folder of instructions, scripts and templates. Claude only loads a skill when it is relevant to the task. That lets you define editorial guidelines, release checks or report formats once and reuse them.
Both expand what Claude can do, and therefore what can go wrong. Anthropic explicitly advises using skills only from trusted sources and reviewing them like installed software. The same applies to MCP servers.
For commercial offerings (API, Team, Enterprise), the Anthropic Privacy Center states that inputs and outputs are not used for training by default. According to the data retention documentation, Anthropic generally deletes API data within 30 days; eligible customers can get zero data retention agreements, although these do not cover every feature.
The consumer plans (Free, Pro, Max) are governed by separate consumer terms. If you work with customer or personal data, use a commercial account, sign a data processing agreement and decide up front which data may be sent to an AI model at all.
The API is billed per token, with separate rates for input and output. According to the pricing page, the range per million tokens runs from US$1 input and US$5 output for Haiku 4.5 up to US$10 input and US$50 output for Fable 5.1. The Batch API halves the cost for work that does not need an immediate result, and prompt caching significantly reduces the cost of context you send repeatedly.
In practice, architecture matters more than list price: which model does each task actually need, how much context is really sent, and which steps can be solved deterministically without a model?
BIT62 helps you use Claude where it makes sense: from choosing the right access path to API integrations, MCP servers and skills, through to Claude Code in your development team, secured by clear permissions, data protection and traceable review steps.
From fast Haiku to thorough Fable, depending on task and budget.
Integrate directly or through Bedrock, Vertex AI and Foundry.
Work agentically in the repository: read, edit, test, commit.
Connect data sources and tools through an open standard.
Provide recurring know-how as reviewable packages.
Commercial access without training on your data.
BIT62 helps with model selection, API integration, MCP servers, skills and Claude Code in your team, with clear permissions and data protection.