Claude Code Chat — VS Code Extension
Bring Claude Code's chat interface from the terminal into VS Code with a beautiful, intuitive UI.
- Source repo
- andrepimenta/claude-code-chat
- Stars
- ★ 1.1k
- Last updated
- 1mo ago
- License
- NOASSERTION
- Primary language
- JavaScript
- FA score
- 48/100 · Major gaps
At a glance
- How it runs
- Works with
- Platform-specificClaude Code
- Cost
- Free software; you pay for model usage
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- Developers who want to access Claude Code without leaving the editor and prefer a graphical interface
- Not a fit if
- Users who don't want to install the Claude Code CLI and configure API keys
- Developers who don't use VS Code or Cursor
- Source review
- 48/100 · Major gaps
What does this agent do, and when should you use it?
Claude Code Chat is a VS Code extension that provides a native graphical chat interface for the Claude Code CLI, eliminating the need to type commands in a terminal. It supports real-time streaming responses, conversation history, Git-based checkpoint restoration, an inline diff viewer, an MCP/Skills/plugin marketplace, image paste with previews, an advanced permissions system, sidebar integration, model selection (including 150+ OpenCredits models), and slash commands. The extension was built using Claude Code itself and requires the Claude Code CLI and an active Claude API or subscription. It works in VS Code, Cursor, and supports Windows/WSL.
The extension listens for chat input in VS Code, forwards messages to the Claude Code CLI (through shell invocation), and displays streaming responses in real time. It provides message copying, syntax-highlighted code blocks, and an auto-resizing input. It integrates with Git to create automatic checkpoints and allow instant restoration to any prior state. It provides an inline diff viewer showing changes in Edit, MultiEdit, and Write messages, with a one-click button to open the native VS Code diff editor. It connects to MCP registries (add-mcp and Anthropic) to browse and install MCP servers, and supports installing skills from skills.sh. It allows pasting images and attaching them to messages (converted to base64 for sending). The permissions system allows or denies tool execution based on user configuration, including a YOLO mode. It supports slash commands (/agents, /cost, etc.), plan mode, and thinking modes (Think/Think Hard/Ultrathink). All functionality is exposed through VS Code commands and the status bar.
- Developers who want to access Claude Code without leaving the editor and prefer a graphical interface
- Developers who want to create checkpoints before experimental refactors or code changes for easy rollback
- Developers who want to see file diffs within messages and review them using VS Code's native diff editor
- Developers who want to browse and install MCP servers, skills, or plugins to extend Claude Code's capabilities
- Security-conscious developers who need granular control over which tools can execute (e.g., Bash, file operations)
- Windows users running Claude Code in WSL who want to integrate it into VS Code
How do you install or deploy this agent?
Prerequisites: VS Code 1.80+ (or Cursor), Claude Code CLI (install from Anthropic), an active Claude API or Pro/Max subscription, and Node.js 18+. In VS Code, run ext install claude-code-chat, or download the .vsix file from releases and run code --install-extension claude-code-chat-x.x.x.vsix.
How do you use this agent?
After installation, press Ctrl+Shift+C (or Cmd+Shift+C on Mac), click the Claude icon in the status bar, or use the Command Palette: Claude Code: Open Chat. Type a message in the chat box and press Enter to send. Use @ to reference files and / to open the slash commands modal. Toggle plan or thinking modes with the switches above the text input. For WSL, configure settings like claudeCodeChat.wsl.enabled and claudeCodeChat.wsl.distro in settings.json.
What are this agent's strengths and limitations?
- Provides a polished native chat interface without needing the terminal
- Git-based checkpoints allow safe rollback to any previous state with one click
- Integrated MCP, skills, and plugin marketplace makes extending functionality easy
- Supports pasting images with previews before sending
- Works in VS Code and Cursor, with native WSL support
- Requires Claude Code CLI and a Claude API/subscription, creating platform-specific dependency
- Installation and setup involve multiple prerequisites (VS Code, Node.js 18+, CLI)
- Full functionality requires network access and API usage, which incurs token costs
- Some claimed features (e.g., full support for 150+ OpenCredits models) are not independently verifiable from the source and may need real-world testing
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Claude Code Chat — VS Code Extension This agent | 48 · Major gaps | IDE extensionFree + model costs | ★ 1.1k | 1mo ago | JavaScript | Claude Code |
| OpenClaude — Universal Terminal AI Coding Agent | 42 · Major gaps | CLIFree + model costs | ★ 34k | today | TypeScript | Claude Code · OpenAI API · Claude API |
| Ogcode | 46 · Major gaps | CLIFree + model costs | ★ 143 | 1d ago | Go | OpenAI API · Claude API |
| agnix Config Linter | 89 · Good | CLIFree | ★ 455 | today | Rust | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows: The extension provides a permission system including interactive permission dialogs, always-allow functionality, YOLO mode, and workspace permissions, but YOLO mode is off by default and permission control relies on user configuration. User confirmation mechanisms exist but are not detailed. Data flow transparency: README mentions real-time cost/token tracking but does not clarify data flow. Sensitive data handling: Supports environment variable passing but does not specify encryption or storage. Dependency security: Uses npm dependencies but no audit or lock file. External effects: Extension executes commands and file operations, but permission controls can limit. Rollback: Provides checkpoint system to restore previous states, evidence is sufficient. Source attribution: README claims built by Claude Code but no verification. Deductions: Permission system exists but default configuration may be permissive, data flow and sensitive data handling lack details, dependency security not audited.
Evidence shows: Code structure is consistent, tests cover downloader functionality, but main extension functionality has only a sample test. Dependency availability: Depends on npm registry and Anthropic CDN, but no offline or fallback plan. Failure messages: Downloader error classification is clear (NETWORK, INTEGRITY, etc.), but other parts of extension error handling not detailed. Deductions: Main functionality tests insufficient, reliance on external services may affect reliability.
Evidence shows: Target users are VS Code users, supports Windows/WSL, offers multiple model choices. Capability boundaries: Permission system can limit tool execution, but YOLO mode may bypass. Trigger precision: Keyboard shortcuts and commands are clear. Environment fit: Supports VS Code and Cursor, but no native Linux support mentioned. Deductions: Capability boundaries rely on user configuration, YOLO mode may reduce security.
Evidence shows: README structure is clear, installation instructions detailed, naming consistent, usage examples provided. Known limitations: Not explicitly listed. License: Custom license with restrictions, not open source. Versioning: Has version number but no CHANGELOG. Maintenance responsibility: Contribution guide exists but maintainer not explicit. Deductions: License is restrictive, missing CHANGELOG and known limitations.
Evidence shows: Output usability is high, provides rich UI features. Marginal value: Provides terminal alternative, unique features like checkpoints. Cost-benefit: Requires Claude API or subscription, but no cost estimation. Deductions: Cost-benefit not quantified.
Evidence shows: README feature claims lack code references, tests cover some functionality. Cross-source verification: No external verification. Fact-inference separation: Some claims in README are inferences. Deductions: Claims lack traceability, test coverage limited.
- The extension downloads and executes the Claude Code binary; ensure the source is trusted.
- YOLO mode skips all permission checks; use with caution.
- The custom license restricts usage; ensure it meets your needs.
- Relies on external npm and CDN; may not work without network.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
🚀 Features Added - New "Fable" tier: A fourth capability tier above Opus, wired end to end — a native Claude Fable card, a fable model for every bundled provider, the ANTHROPIC_DEFAULT_FABLE_MODEL environment variable and its removal path, and local-router mapping. Capability order is now Fable → Opus → Sonnet → Haiku. …View on GitHub ↗ -
Release notes excerpt
🚀 Features Added - Sponsor button: Added a Sponsor button on both the GitHub repository and the VS Code Marketplace listing, linking to GitHub Sponsors. - GPT 5.5 as the default GPT model: The "GPT" quick-select now targets OpenAI's mainline GPT 5.5, replacing the previous GPT 5.3 Codex entry, with a faster gpt-5.4-mini tier. 🐛 Bug Fixes - Valid fast-tier GPT fallback: The bundled model fallback no longer points the GPT fast (haiku) tier at a non-existent openai/gpt-5.5-mini. It now uses openai/gpt-5.4-mini, matching what the live resolver selects. …View on GitHub ↗ -
Release notes excerpt
[2.0.6] - 2026-04-23 🚀 Features Added - Smarter post-install setup: Fresh installs now "just work" without a VS Code restart. After install, the extension checks whether claude resolved on your PATH and, if not, auto-configures claudeCodeChat.executable.path to the known install location. An existing custom executable path is respected. …View on GitHub ↗