Feature availability

The local CLI, and everything that runs on your own machine, works with every provider. The differences show up in features that depend on a claude.ai account or on a server-side capability that only some providers expose. This page lays those differences out so you can check before you commit to a provider, or work out why a feature is missing in your setup.

Key used in the tables: Yes means available, No means not available, and anything else is a qualified yes. "Admin-enabled" means off until an organisation admin turns it on.

Finding your column

Your column depends on how Claude Code authenticates:

Works everywhere

• The CLI and the Agent SDK
• The VS Code and JetBrains extensions
• Subagents, hooks, commands and skills
• CLAUDE.md memory, plugins and MCP servers
• Checkpoints, sandboxing and workflows
• OpenTelemetry metrics and the managed settings file

A few of those have provider-specific wrinkles:

• MCP. claude.ai connectors only load when your claude.ai subscription is the active credential. Tool search is off by default when ANTHROPIC_BASE_URL points somewhere other than Anthropic, and is not supported on Vertex AI models older than the Claude 4.5 generation or on Foundry deployments hosted on Azure.
• The Explore subagent. When the main conversation runs Fable, Explore runs on Opus for subscription, Console and ANTHROPIC_BASE_URL gateway users. Everywhere else, including Claude Platform on AWS, it runs on Fable.
• Commands. /design-sync and /import (and claude import) are unavailable on Bedrock, Vertex AI, Foundry, Claude Platform on AWS and through a Claude apps gateway. /voice needs a claude.ai account. /list-agents and its alias /peers only appear where cross-session messaging is enabled.

Needs a Claude subscription

These require a claude.ai login. A Console key or cloud credentials will not reach them:

• Cloud sessions and Claude Code on mobile
• Claude Code in Slack (Pro and Max)
• Claude Code Desktop
• Routines (/schedule)
• Ultrareview
• Code Review (Team and Enterprise)
• Remote Control
• The Chrome extension
• Computer use (Pro and Max)
• Artifacts (Pro, Max, Team, Enterprise)
• Voice dictation

Desktop is a partial exception. Gateway routing can be set in the app or by an admin, Enterprise can route Desktop to Vertex AI or a gateway provider through Desktop's managed configuration, and "Claude Desktop on 3P" runs the Code tab against Bedrock, Vertex AI, Foundry or a self-hosted LLM gateway.

Server-side capabilities by provider

Auto mode on Bedrock, Vertex AI and Foundry supports only Claude Sonnet 5 or later, Opus 4.7 or later, Haiku 5.5 and the Fable models. Versions v2.1.158 to v2.1.206 also needed CLAUDE_CODE_ENABLE_AUTO_MODE=1 on these providers; v2.1.207 dropped that requirement.

Cross-session messaging needs v2.1.224+ on macOS and Linux (including WSL 2) or v2.1.234+ on native Windows. With API keys it is same-machine only. On Bedrock, Claude Platform on AWS, Vertex AI and Foundry it is same-machine only and needs v2.1.248+. Reaching cloud sessions or other machines requires a session connected to Remote Control, which in turn needs a claude.ai sign-in.

Admin and analytics by provider

Gateways

Behind an LLM gateway, you get whatever the underlying provider supports, minus anything Claude Code switches off itself. Whenever ANTHROPIC_BASE_URL points at a host other than api.anthropic.com, Remote Control and server-managed settings are disabled regardless of where the gateway forwards. Anthropic-only features such as the Advisor work only if the gateway passes requests through to the Anthropic API unchanged. The gateway protocol page explains how client behaviour differs between Bedrock- or Vertex-format gateways, ANTHROPIC_BASE_URL gateways and Claude apps gateway sign-ins.

What each provider is missing

Anything not listed below behaves as it does on a subscription. On Bedrock, Vertex AI, Foundry and Claude Platform on AWS, error reporting and telemetry to Anthropic are off by default; see data usage.

Amazon Bedrock

• Missing: every subscription-only feature, plus web search, fast mode, Advisor, channels, the analytics dashboard, server-managed settings, /design-sync and /import.
• Partial: Desktop only through Claude Desktop on 3P; auto mode on newer models only; cross-session messaging on the same machine; zero data retention per your AWS agreement.
• Workarounds: use /loop instead of /schedule; use GitHub Actions or GitLab CI/CD instead of cloud sessions; use WebFetch with a specific URL instead of web search.

Claude Platform on AWS

• Missing: every subscription-only feature, plus fast mode, Advisor, channels, GitHub Actions, the analytics dashboard, server-managed settings, /design-sync and /import.
• Has, unlike Bedrock: web search.
• Partial: cross-session messaging on the same machine.
• Workarounds: /loop for scheduling; GitLab CI/CD for unattended runs.

Google Vertex AI

• Missing: every subscription-only feature, plus fast mode, Advisor, channels, the analytics dashboard, server-managed settings, /design-sync and /import.
• Partial: Desktop via Desktop managed configuration or Claude Desktop on 3P; web search on Claude 4 and later; auto mode on newer models only; same-machine cross-session messaging; zero data retention per your Google Cloud agreement.
• Workarounds: /loop; GitHub Actions or GitLab CI/CD.

Microsoft Foundry

• Missing: every subscription-only feature, plus fast mode, Advisor, channels, GitLab CI/CD, the analytics dashboard, server-managed settings, /design-sync and /import.
• Partial: Desktop only through Claude Desktop on 3P; auto mode on newer models only; same-machine cross-session messaging; zero data retention per your Azure agreement.
• Workarounds: /loop; GitHub Actions.

Anthropic Console

• Missing: every subscription-only feature.
• Everything in the server-side capabilities table is available, though fast mode needs provisioned access. Server-managed settings work when the key belongs to a Team or Enterprise organisation.

By subscription plan

This only applies when you sign in with a claude.ai account. HIPAA-configured Enterprise organisations have some of these turned off (see HIPAA setup).

Models

Which models and context sizes you can use depends on provider and region; see model configuration. Vision, PDF input and extended thinking are properties of the model, so they work wherever the model is offered. Prompt caching behaves the same on most providers, but on Bedrock support varies by model.

Related

• Enterprise deployment options
• Amazon Bedrock
• Google Vertex AI
• Microsoft Foundry
• Platforms