A Go port of Claude Code — Anthropic's agentic CLI coding assistant
Fast. Extensible. Built for terminals.

claw-code-go is a full Go reimplementation of Claude Code — Anthropic's agentic coding assistant. It runs in your terminal, understands your codebase, calls tools, writes and edits files, searches the web, and works autonomously until the job is done.

Why Go?

• Single static binary — no Node.js runtime required
• Faster startup, lower memory footprint
• Easier to embed, cross-compile, and distribute

Full tool-use conversation loop: Claude reasons, calls tools, reads the results, and keeps going until the task is complete or it hits an end_turn.

Rich terminal UI built with Bubble Tea — streaming output with a spinner, styled message history, syntax-aware theming, and a clean session view.

Native OAuth flow for Anthropic accounts. Credentials are stored securely and refreshed automatically. OpenAI is scaffolded alongside Anthropic — swap providers via config.

Full MCP integration: connect external tool servers over stdio or SSE, register their tools, and let Claude call them seamlessly alongside built-in tools.

Fine-grained permission system controls what Claude can do — bash execution, file writes, network access — with configurable modes (auto, ask, deny) and rule-based overrides.

On every turn, claw-code-go automatically injects:

• Git repo state (branch, diff summary, recent commits)
• Memory directory files (.claw-code/memory/)
• System info (OS, hostname, working directory)

Sessions are saved as JSON files in ~/.claw-code/sessions/. Resume any previous session by ID. The TUI includes a built-in session history browser.

Every session tracks input/output tokens per turn. Estimated USD cost is shown at the end of each session for all known models (Claude 3/4 + GPT-4o family).

Automatic conversation compaction keeps context windows manageable on long sessions — summarising older turns without losing important state.

Drop markdown files into .claw-code/memory/ and they're injected into every conversation automatically — persistent instructions, project context, preferences.

• Go 1.24+
• ANTHROPIC_API_KEY environment variable (or log in via --login)

git clone https://github.com/daolmedo/claw-code-go
cd claw-code-go
go build -o claw-code-go ./cmd/claw-code-go

Or install directly:

go install github.com/daolmedo/claw-code-go/cmd/claw-code-go@latest

./claw-code-go

export ANTHROPIC_API_KEY=sk-ant-...
./claw-code-go --prompt "Refactor the auth package to use interfaces"

./claw-code-go --login
# or explicitly:
./claw-code-go --login --provider anthropic

./claw-code-go --login --provider openai
# Prompts for your OpenAI API key and stores it securely

Credentials are saved to ~/.claw-code/credentials/ and reused automatically on the next run. Switch providers at any time with --provider.

./claw-code-go --session <session-id>

claw-code-go/
├── cmd/claw-code-go/        # CLI entry point
└── internal/
    ├── api/                 # Anthropic API client (SSE streaming, types)
    ├── auth/                # OAuth flow, credential storage, token refresh
    ├── commands/            # Slash command registry
    ├── compat/              # Upstream TS source parity manifest
    ├── config/              # Config loading, permission modes, rules
    ├── context/             # Context assembly (git, memory, sysinfo)
    ├── mcp/                 # Model Context Protocol client (stdio + SSE)
    ├── permissions/         # Permission enforcement & rule engine
    ├── runtime/             # Agentic conversation loop, session persistence
    ├── tools/               # Built-in tool implementations
    ├── tui/                 # Bubble Tea TUI (model, styles, theme, history)
    └── usage/               # Token tracking and cost estimation

• Phase 1 — Foundation: API client, basic conversation loop, core tools
• Phase 2 — TUI: Bubble Tea UI, streaming, spinner, slash commands
• Phase 3 — OAuth + multi-provider scaffolding
• Phase 4 — MCP (Model Context Protocol) integration
• Phase 5 — Permissions & Safety
• Phase 6 — Context compaction
• Phase 7 — Multi-provider login UX
• Phase 8 — Compat harness modes
• Phase 9 — Full TUI foundation (themes, history view)
• Phase 10 — Core tool expansion (file_edit, web_fetch, web_search, todo_write)
• Phase 11 — Permissions + config system
• Phase 12 — Context assembly + memory directory
• Phase 13 — Cost tracking + session history UI
• Phase 14 — LSP integration
• Phase 15 — Plugin/extension system

PRs welcome. Run go build ./... and go vet ./... before submitting.

MIT