Claude Code Timeline: Release Date and Major Updates (2026)

Claude Code is Anthropic’s AI coding agent for working inside your project. Anthropic released it in research preview on February 24, 2025, and made it generally available on May 22, 2025.

The following timeline records the public milestones that changed how you can use Claude Code, including its terminal release, IDE support, project configuration, cloud and remote sessions, automation, and agent coordination.

It uses the date Anthropic announced or released each capability that materially changed how people use Claude Code. Routine numbered changelog entries stay outside the timeline.

Note that Claude Code has a separate history from Claude models and Anthropic’s other apps. This page includes a model only when it changed Claude Code. See the Claude Timeline for the wider model history.

Last updated: September 1, 2026.

Claude Code timeline at a glance

2024: MCP creates the integration foundation

November 25, 2024: Anthropic announces MCP

Model Context Protocol, usually called MCP, was announced as an open standard for connecting AI applications to tools and data sources. Claude Code launched later and adopted MCP as one of its main extension paths. You can use MCP to give a coding session access to a ticket, an error report, a database, a design file, or a private service without pasting all of that context into a prompt.

Explore Model Context Protocol servers for development and agent workflows.

2025: from terminal preview to a broader coding product

February 24, 2025: Claude Code enters research preview

Anthropic introduced Claude Code as a command-line tool for agentic coding on the same day it released Claude 3.7 Sonnet. The preview could inspect a codebase, edit files, run commands and tests, use Git, and complete tasks through a terminal session. The initial framing mattered: Claude Code was built to act inside a real development environment, not to produce isolated snippets in a chat window.

May 22, 2025: general availability and Claude 4

On May 22, Anthropic moved Claude Code out of research preview. The same announcement introduced Claude Opus 4 and Claude Sonnet 4, two models designed for difficult coding, reasoning, and agent work. Anthropic also announced native integrations for Visual Studio Code and JetBrains IDEs, plus GitHub Actions support for background tasks. From this point, you could use Claude Code beyond a terminal-only workflow.

June 18, 2025: remote MCP servers arrive

Remote MCP support let Claude Code connect to hosted servers through a URL and OAuth authentication. You no longer had to run every integration locally. Connect tools such as Linear or Sentry when you need issue details, project context, and error reports in the same session that edits the repository. Keep local MCP servers for tools that must run only on your machine.

June 30, 2025: Hooks become available

Hooks added event-driven control around Claude Code actions. Your project can run a command, a prompt, or another check when a session starts, before a tool runs, after a tool completes, or when Claude stops. Use them for project habits that should not depend on the model remembering an instruction, such as formatting after edits, checking a command against a policy, recording a session summary, or blocking an unsafe operation.

July 24 to July 30, 2025: custom Subagents and agent controls

The July releases opened a more explicit delegation model. You could define Custom Subagents through /agents, then call one for a focused task such as code review, investigation, testing, or documentation. Subsequent releases added SessionStart Hooks, direct mentions for custom agents, and model selection for agents. Keep the main session responsible for the task while an isolated agent explores part of the codebase and returns a summary.

August 5 and August 20, 2025: Opus 4.1 and business controls

Claude Opus 4.1 arrived on August 5 with an emphasis on coding precision and agent tasks. Two weeks later, Anthropic announced Claude Code access and administration for Team and Enterprise plans. If you manage a workspace, you could assign premium seats that included Claude Code and apply administrative controls. Claude Code then became part of the same subscription and governance decisions as the broader Claude product.

September 29, 2025: Claude Sonnet 4.5

Claude Sonnet 4.5 became a new model option for agents, coding, and computer use. It belongs in this timeline because a model release changes the practical choice you make inside a Claude Code session. The model does not replace your tools, project instructions, permissions, or integrations. It changes the reasoning and coding capability that works through them.

October 20, 2025: Claude Code on the web

Claude Code on the web entered research preview for Pro and Max users, with a later expansion to eligible Team and Enterprise users. After you connect a GitHub repository, it runs tasks on Anthropic-managed cloud infrastructure. Each task receives an isolated environment, can continue after you close the browser, and can create a pull request for review. Choose web sessions for work that can proceed with less continuous steering than a local terminal session.

October 27 and October 31, 2025: Plan Mode and Plugins

The October 27 release gave Plan Mode a dedicated Plan subagent. In Plan Mode, you can inspect a repository and ask for an approach without editing source files or running commands that change the project. Four days later, Anthropic announced Plugins and plugin marketplaces. A plugin can package commands, agents, Skills, Hooks, MCP servers, and related configuration so you can share a tested setup across projects or with colleagues.

November and December 2025: cloud handoff, Opus 4.5, and Slack

November releases expanded the handoff between local and cloud work. You could send background work to the web, and custom agents gained more configuration options. Claude Opus 4.5 followed on November 24 for demanding coding and agent tasks. On December 8, Claude Code in Slack entered research preview. Mention @Claude in a Slack channel to send the surrounding discussion into a Claude Code web session, then receive progress updates and a pull request link in the same thread.

2026: multi-agent work, remote sessions, and automation

February 5, 2026: Claude Opus 4.6 and agent teams

Claude Opus 4.6 added a new high-capability model for planning, code review, debugging, and longer tasks. The release also introduced agent teams in research preview. Use agent teams when you need several independent Claude Code sessions to work on related tasks at the same time. They can share work through the team structure and communicate beyond a single parent conversation.

February 20, 2026: desktop review tools and Claude Code Security

Claude Code Desktop gained tools for previewing a running app, reviewing pull requests, responding to CI failures, and moving a session between Desktop, web, mobile, and CLI. That same day, Anthropic announced Claude Code Security as a limited research preview for Team and Enterprise customers. It scans codebases for vulnerabilities and proposes patches for your review. Treat it as a specialized product surface, not a replacement for ordinary code review or security testing.

February 25 to March 19, 2026: Remote Control, memory, and Channels

Remote Control expanded to more users in late February. It connects your local Claude Code session to claude.ai/code or the Claude mobile app while execution and filesystem access remain on your machine. Claude Code also added auto memory for useful session context and later added names for Remote Control sessions. On March 19, the --channels preview opened a different path: selected MCP servers could push events into an active session.

March 24 and July 10, 2026: Auto Mode moves from preview to general availability

Auto Mode began as a Team-plan research preview on March 24. It uses a classifier to examine tool calls before they run. Routine actions can proceed without a separate approval, while risky actions are blocked or returned to you for a decision. Anthropic later updated the announcement to say Auto Mode was generally available on July 10. Your access can still depend on a plan, provider, administrator setting, or workspace policy, so check your Claude Code settings before relying on it.

April 14, 2026: a redesigned Desktop app and Routines

Anthropic released a new Desktop workspace with a sidebar for sessions, a flexible layout, an integrated terminal, and a file editor. It arrived on the same day as Routines, a research-preview feature for repeatable cloud automations. A Routine can run on a schedule, after an API call, or in response to a GitHub event. It keeps its own prompt, repository, and connected tools. Use it when you want managed automation for issue triage, documentation checks, or release verification.

May 11 to June 2, 2026: Agent View, Remote Control in VS Code, and dynamic workflows

Agent View entered research preview on May 11. It lists Claude Code sessions that are running, blocked, or finished, and /goal lets you define a completion condition for persistent work. On May 13, the VS Code extension gained /remote-control. Dynamic workflows arrived in research preview on June 2. Claude Code can create a task-specific harness that splits work among agents, checks results, and combines the output. Use it for complex work where one fixed prompt or one linear agent loop is not enough.

May 28 to July 24, 2026: new models for Claude Code

Claude Opus 4.8 arrived on May 28 for coding and longer agent tasks. Anthropic announced Claude Fable 5 and Claude Mythos 5 on June 9 with different availability and safety policies. Claude Sonnet 5 followed on June 30 for coding, tool use, agents, and professional work. On July 24, Claude Code 2.1.219 made Claude Opus 5 the default Opus model, with a one-million-token context window. Because model names and access can change quickly, focus on which model your account, provider, and task can use. The product’s extension layer remains separate from that choice.

August 4, 2026: Focus view and stronger session isolation

Claude Code 2.1.221 added Focus view to the VS Code extension. The view hides the full stream of tool activity behind an expandable summary for each turn while a live indicator shows when a tool is still running. You can switch it from the chat menu, use Ctrl+Alt+F, or run the Claude Code: Toggle Focus view command. The ordinary transcript remains available when you need to inspect a command or result. This gives long, tool-heavy sessions a quieter editor view without removing access to their execution details.

The same release tightened how isolated work interacts with repositories and credentials. Sessions created with /fork now receive their own Git worktree, which keeps the fork’s checkout separate from the original session. Background sessions were also told to preserve completed work through commits and pushes, respect the repository’s CLAUDE.md Git rules, and open a draft pull request only when the task calls for one. These defaults make the destination of unattended work clearer, but you should still review the branch, commit, and remote before merging.

For Linux and WSL sandboxes, credential-file rules gained mode: "mask". A sandboxed command reads a sentinel value from the protected file, while the sandbox proxy substitutes the real secret when matching traffic leaves the environment. An optional extraction pattern can limit masking to selected spans. On macOS, the same setting falls back to denying access because file masking is not available there. This control still requires a narrow destination list, credential rotation, and review of commands that can send data outside the project.

August 7, 2026: self-hosted environments and cross-session messaging

Claude Code 2.1.224 added self-hosted environments for Team and Enterprise plans. Anthropic announced the feature as a public beta; it is off by default and unavailable to organizations using Zero Data Retention (ZDR). The claude self-hosted-runner command turns a machine or container you control into an execution environment for Claude Code sessions started from the web, mobile app, or Desktop. This expands cloud-style delegation to organizations that need tasks to run on their own infrastructure. The runner does not make every local setup suitable for unattended work: administrators still need to control repository access, network destinations, credentials, isolation, updates, and the lifecycle of the host.

The same release added cross-session SendMessage and ListAgents on macOS and Linux. Sessions can discover and message other Claude Code sessions across machines, with approval settings for inbound messages sent to sessions that bypass normal permissions. Claude Code also gained an archive plugin source, which installs a plugin from a ZIP file over HTTPS and can pin the archive to a SHA-256 hash. Teams can now choose where sessions run, coordinate separate sessions across machines, and distribute internal plugin packages without a Git or npm source.

August 13 and 14, 2026: default subagent forking and deeper GitLab support

Claude Code 2.1.232 made subagent forking the default. A forked subagent with subagent_type: "fork" inherits the full conversation and prompt cache, so it starts with the parent session’s working context. Non-teammate agents launched from an interactive session now run in the background by default. The main session can continue while a focused agent works with the same history. Forks can consume more context and tokens than isolated subagents, so use them for work that depends on the current conversation.

The release also made cross-session coordination easier. Type @ and a live session name to send it a message through SendMessage, then choose whether inbound messages are accepted, held, or refused. GitLab support expanded to plugin marketplaces and common token formats. Claude Code 2.1.233 then added GitLab merge request URLs to --worktree and claude agents. GitLab teams can install marketplace sources, protect token values, and connect isolated worktree sessions to merge requests. Use unique session names and keep approval policies around sessions with broad permissions.

August 20, 2026: authenticated plugin sources and self-hosted runner controls

Claude Code 2.1.238 added authenticated plugin sources. A URL marketplace or catalog entry can use a headersHelper command to mint HTTP headers, such as a short-lived token, for catalog and same-origin archive fetches. Claude Code shows the command before a catalog plugin install or update and asks for confirmation unless you pass -y.

Self-hosted runners gained --defer-shutdown-max-min to keep attached sessions alive, park remaining work after a grace period, and then exit. --proxy-authorization-command and --proxy-authorization-file can supply fresh Proxy-Authorization headers for each outbound connection. These options give Team and Enterprise operators more control over runner shutdown and proxy-based egress.

August 27, 2026: restricted mode for controlled evaluation

Claude Code 2.1.248 added --restricted for evaluation harnesses on shared machines. It removes built-in command and code execution tools plus WebFetch unless you name them explicitly with --tools. It confines file tools to the working directories, ignores user, project, and local settings, and refuses bypassPermissions. Managed settings and values passed through --settings still apply. Evaluation operators get a repeatable boundary independent of personal host configuration.

The same release extended same-machine SendMessage and ListAgents support to Bedrock, Vertex AI, Foundry, and sessions with telemetry disabled. Those environments can now use the cross-session coordination introduced earlier in August.

September 1, 2026: Claude Fable 5.1 and tighter Auto Mode boundaries

Claude Code 2.1.257 added Claude Fable 5.1 as the default Fable model with a one-million-token context window. Auto Mode no longer automatically approves requests for cloud metadata credentials, attempts to evade network controls, or cross-tenant access unless the environment expects them. Before its first read outside the working directories, it asks once and can block later external reads through permissions.blockReadsOutsideWorkingDirectories.

How Claude Code customization fits together

CLAUDE.md is your project context file. Claude Code reads it at the start of a session. Put repository commands, architecture notes, coding conventions, test requirements, and review rules there. A root file can set broad project guidance, while files in subdirectories can carry local instructions for a package or service. Use it for context that should apply repeatedly.

Hooks respond to events. Use them when a command or policy must run predictably, such as formatting after a write, checking a changed file, or preventing a prohibited action. A Hook does not depend on Claude choosing to follow a paragraph of guidance. That makes it suitable for safeguards and mechanical project tasks.

MCP servers connect Claude Code to external systems. A local server can expose a tool on your machine. A remote server can connect through a hosted URL and OAuth. Use MCP when your task requires issue data, logs, design context, internal documents, or a company service that Claude Code cannot reach through the repository alone.

Skills package reusable instructions and workflows. You can invoke a Skill directly, or Claude can load one when its description matches the task. Choose a focused Skill when a workflow applies to code review, release preparation, a framework, or a deployment process. Keep the global instruction file for guidance that belongs in most sessions.

See the Best Agent Skills guide for examples beyond Claude Code.

Plugins bundle capabilities for distribution. A plugin can include Skills, Subagents, Hooks, MCP configuration, and commands. Use one to share a tested setup without asking every developer to recreate it by hand. Marketplaces give your project or organization a place to discover and manage those packages.

Subagents and agent teams divide work. A Subagent has its own context and returns a result to your main session. Agent teams coordinate several independent sessions for work you can split, such as a large codebase review, separate investigations, or competing implementation proposals. Set clear task boundaries so agents do not duplicate work or touch the same files.

Plan Mode is the safer place to inspect a repository and agree on an approach before edits begin. It can read files and explain a proposed change, but it does not edit the project. Use it when a task is ambiguous, has a broad blast radius, or needs a design decision before implementation.

Model changes that affected Claude Code

Claude Code launched with Claude 3.7 Sonnet. The terminal agent supplied repository access, tools, and an approval flow so the model could work inside a development environment.

Claude Opus 4 and Claude Sonnet 4 arrived with general availability in May 2025. Opus 4.1, Sonnet 4.5, and Opus 4.5 later expanded the choices. Repository context and tool permissions still determine what each model can inspect and change.

The 2026 releases added Opus 4.6, Opus 4.8, Fable 5, Mythos 5, Sonnet 5, and Opus 5. Fable 5.1 became the default Fable model in September.

If you need the complete model sequence, use the Claude Timeline.

Model access depends on your subscription, API account, provider, and gateway configuration. Check the model selector and current account documentation before choosing one for a task.

Commands and configuration changes that shaped the workflow

Claude Code starts in a project directory with claude. Use /init for CLAUDE.md, /agents for Custom Subagents, /mcp for connections, and /plugin for plugin management.

Plan Mode provides a read-only path for investigation. Accept Edits reduces file-edit prompts, while Auto Mode uses classifier-based screening for eligible accounts and environments. The older --dangerously-skip-permissions option lacks the same safeguards. Projects with production credentials or destructive scripts still need explicit permissions and Hooks.

Sandbox rules constrain commands and network access. Credential masking keeps selected secret files out of command context on Linux and WSL while a proxy supplies the real value to approved destinations. Restricted mode removes command execution and WebFetch by default, confines file tools to working directories, and excludes personal project configuration.

Remote work added more commands. /remote-control connects a local session to a browser or mobile device, /schedule maps repeatable work to Routines, and /goal defines a completion condition. Agent View shows work that is running or waiting for input. Review diffs, tests, and pull requests as these features run.

Configuration also became more layered. Keep shared instructions in CLAUDE.md, personal settings in your user directory, and local exceptions in ignored settings files. Managed settings apply organization rules, while Plugins distribute Skills, Hooks, agents, and MCP configuration.

For syntax and a command-by-command reference, use the Claude Code Commands Cheat Sheet.

What changed over time

Claude Code began as a terminal agent for local repository tasks. Its first year added model choices, IDE views, remote MCP, Hooks, project instructions, Custom Subagents, enterprise controls, and Plugins.

Later releases let work move between terminal, editor, Desktop, browser, phone, and Slack. Routines, Agent View, Auto Mode, agent teams, and dynamic workflows support longer tasks. Restricted mode gave evaluation harnesses a fixed tool boundary.

FAQs

When was Claude Code released?

Anthropic introduced Claude Code on February 24, 2025, as a limited research preview with Claude 3.7 Sonnet. It became generally available on May 22, 2025.

Is Claude Code free?

No. The free Claude.ai plan does not include Claude Code. To use it regularly, you need a paid Claude plan, such as Pro, Max, Team, or Enterprise, or a billed developer account through Anthropic Console, Amazon Bedrock, Google Vertex AI, or Microsoft Foundry. Subscription use counts toward your plan’s usage limits. API and cloud-provider use is billed by tokens.

Which Claude model should I use with Claude Code?

Start with default unless you have a reason to override it. It selects Anthropic’s recommended model for your account. For everyday implementation, use sonnet. Choose opus for difficult debugging, architecture decisions, code review, or a high-risk multi-file change. Choose opusplan when you want Opus to make the plan and Sonnet to carry it out. Use haiku for simple, fast, lower-cost tasks. Run /model to switch during a session.

What is the difference between Claude Code on the web and Remote Control?

Claude Code on the web runs a task in Anthropic-managed cloud infrastructure. Remote Control lets your browser or mobile device steer a Claude Code session that is still running on your own machine. Use Remote Control when local files, tools, or MCP servers are part of the task.

What should go in CLAUDE.md?

Put stable project guidance in CLAUDE.md: how to run tests, coding conventions, architecture decisions, preferred libraries, repository rules, and review requirements. Keep reusable task workflows in Skills, automatic checks in Hooks, and external-service access in MCP configuration.

Are Claude Code Plugins the same as Skills?

No. A Skill is a reusable set of instructions or workflow guidance. A Plugin is a distributable package that can include Skills alongside commands, Subagents, Hooks, MCP servers, and other Claude Code components.

What is Plan Mode in Claude Code?

Plan Mode is a read-only working mode for investigating a codebase and proposing a change before implementation. Claude Code can inspect files and explain an approach, but it does not edit files or run commands that change the project. Use it when you need to review the plan before granting execution access.

More Related Resources

• Most Popular Agent Skills on GitHub for Coding Agents
• The Ultimate Claude Code Resource List: Agents, Skills, Plugins & More
• Claude Code Pricing and Usage Limits: Free, Pro, Max, Team, and API
• 7 Best CLI AI Coding Agents
• Codex Timeline: Release Dates and Major Updates
• Free AI Tools for Developers
• Claude Code MCP Setup: Add Local and Remote MCP Servers