open-claude-code
github.com/ruvnet/open-claude-code- Category
- Developer Tools
- Rank
- No. 1887Tools index
Previous survey · No. 1892 ·
- Pricing
- Open Source
- Type
- TOOL
- Builder
- ruvnet
- GitHub
- 494 stars
- Latest release
- v2.0.0-nightly.20260906
- Date
About
rUv's nightly decompile and reverse-engineered rebuild of the Claude Code CLI — open-source attempt to recreate the proprietary tool.
What it does
Open Claude Code is a terminal coding agent that sends prompts to supported model providers, streams responses, and lets the model inspect or change a project through registered tools. It offers one-shot commands and an interactive terminal interface, with sessions, permissions, hooks, custom agents, skills, context compaction, and MCP connections.
Why it's ranked here
The scope is unusually broad, but the repository gives mixed signals about maturity. The entry point wires together a real agent loop, tool registry, terminal interface, sessions, hooks, and MCP clients. However, one architecture document claims complete parity while another records substantial robustness, security, interface, and advanced-feature gaps. Some shipped tools are explicitly stubs.
What's good
The command line supports interactive work, plain output, JSON, and streamed JSON, making it useful both for people and scripts. Permission modes, tool allowlists, hooks, project instructions, custom agents, and skills provide meaningful control. MCP support covers four transports, while provider options include Anthropic, OpenAI, Google, Bedrock, and Vertex configurations.
Tradeoffs
Several headline claims exceed the implementation evidence supplied. The language-server tool only provides limited diagnostics and returns placeholder messages for completions, hover, definitions, and references. Project documents also describe earlier gaps in sandboxing, permission prompts, path sanitization, richer file handling, worktree isolation, OAuth, telemetry, cron persistence, and full provider credential flows. Documentation counts for tests, commands, files, and settings conflict.
How to use it well
Use it when you want a configurable Node-based coding agent for terminal experiments, scripted repository queries, or testing MCP and provider integrations. Start with read-only permissions and narrow tool allowlists, then enable mutations after validating behavior on disposable work. It suits engineers willing to inspect implementation details. It does not establish production-grade security equivalence or complete language-server integration.
Technical notes+
v2/src/index.mjs composes the agent loop, registry, permission checker, hooks, MCP clients, sessions, checkpoints, prompt cache, skills, agents, telemetry, and optional optimization cascade. It selects non-interactive output or an Ink interface, then falls back to readline. v2/src/ui/app.mjs implements the React terminal view and keeps a bounded message history; v2/src/ui/repl.mjs provides the fallback REPL. v2/src/tools/lsp.mjs explicitly labels most LSP operations as stubs, while diagnostics invoke TypeScript or Python tooling with argument arrays. scripts/decompile-and-diff.mjs compares decompiled package structures, and scripts/auto-update-capabilities.mjs can ask an Anthropic model to regenerate selected source files before running tests. docs/adr/ADR-002-path-to-100-percent.md documents material implementation gaps that temper stronger parity claims in README.md and docs/adr/ADR-001-v2-architecture.md.
Observed
- License
- MIT
- Primary language
- JavaScript using Node.js ESM modules
- Runtime
- Node.js 22+ is specified in the architecture document
- Packaging
- Scoped npm package with npx, global installation, and source execution options
- Interfaces
- Command-line interface with one-shot, interactive, text, JSON, and streamed JSON modes
- MCP
- Client support is documented for stdio, SSE, WebSocket, and Streamable HTTP transports
- Provider surface
- Configuration examples cover Anthropic, OpenAI, Google, AWS Bedrock, and Google Vertex
- Terminal UI
- Ink and React provide the primary interactive interface, with a readline fallback
Read from README.md, docs/adr/ADR-001-v2-architecture.md, docs/adr/ADR-002-path-to-100-percent.md, docs/adr/ADR-003-nightly-verified-release-pipeline.md, scripts/decompile-and-diff.mjs, scripts/format-diff-summary.py, scripts/auto-update-capabilities.mjs, v2/src/index.mjs, v2/src/ui/app.mjs, v2/src/ui/repl.mjs, v2/src/tools/ls.mjs, v2/src/tools/lsp.mjs.
What it can do
Decompile software binaries
Compiled binary files → Decompiled source code
Reverse engineer proprietary applications
Proprietary software executable → Reconstructed code structure and logic
Rebuild applications from reverse-engineered code
Reverse-engineered code components → Functional rebuilt application
Recreate Claude Code CLI functionality
Code analysis requests → Code analysis and suggestions
Provide open-source alternative to proprietary CLI tool
Command line instructions → Executed CLI operations
Tags
Tech Stack
Media
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.