A Rust coding agent that uses the official Claude Agent SDK (Claude Code) or ChatGPT Pro via OAuth.

• No API keys needed - Uses Claude Code CLI credentials and Codex OAuth
• Dual provider support - Works with Claude Agent SDK and OpenAI/Codex
• Streaming responses - Real-time output as the model generates
• Server/Client architecture - Run as daemon, connect from multiple clients
• 12 built-in tools - File ops, search, web, shell, and parallel execution

You need at least one of:

• Claude Max subscription - Install the SDK: pip install claude-agent-sdk, then run claude to authenticate
• ChatGPT Pro/Plus subscription - Run codex login to authenticate

cargo install --path .

Or build from source:

cargo build --release
./target/release/jcode

# Interactive REPL (default)
jcode

# Run a single command
jcode run "Create a hello world program in Python"

# Start as background server
jcode serve

# Connect to running server
jcode connect

# Specify provider explicitly
jcode --provider claude
jcode --provider openai

# Change working directory
jcode -C /path/to/project

jcode supports MCP servers, allowing you to extend its capabilities with external tools.

Add MCP servers to .claude/mcp.json (project-local) or ~/.claude/mcp.json (global):

{
  "servers": {
    "playwright": {
      "command": "npx",
      "args": ["@anthropic/mcp-playwright"],
      "env": {}
    },
    "filesystem": {
      "command": "/path/to/mcp-filesystem",
      "args": ["--root", "/home/user"],
      "env": {}
    }
  }
}

Servers are automatically connected on startup and their tools are available with the prefix mcp__servername__toolname.

The agent can manage MCP servers at runtime using the mcp tool:

# List connected servers and their tools
{"action": "list"}

# Connect to a new server
{"action": "connect", "server": "my-server", "command": "npx", "args": ["@some/mcp-server"]}

# Disconnect from a server
{"action": "disconnect", "server": "my-server"}

# Reload from config file
{"action": "reload"}

MCP servers communicate via JSON-RPC 2.0 over stdio. See the MCP specification for details on implementing your own servers.

┌─────────────────────────────────────────────┐
│              CLI / Client                   │
├─────────────────────────────────────────────┤
│         Server (Unix Socket)                │
├─────────────────────────────────────────────┤
│              Agent Loop                     │
├─────────────────────────────────────────────┤
│            Provider Trait                   │
│  ┌──────────────┐  ┌──────────────┐        │
│  │ Claude Agent │  │ OpenAI/Codex │        │
│  │     SDK      │  │    OAuth     │        │
│  └──────────────┘  └──────────────┘        │
├─────────────────────────────────────────────┤
│              Tool System                    │
│  bash │ read │ write │ edit │ glob │ ...   │
└─────────────────────────────────────────────┘

J-Code uses Claude Agent SDK to talk to Claude Code. Claude Code credentials are stored at:

• ~/.claude/.credentials.json (Claude Code CLI)
• ~/.local/share/opencode/auth.json (OpenCode, if installed)

OpenAI/Codex OAuth credentials are still stored at:

• ~/.codex/auth.json

For provider/auth details, see OAUTH.md.

• cargo test
• cargo run --bin test_api (Claude Agent SDK smoke test)
• cargo run --bin jcode-harness (tool harness; add --include-network to exercise web tools)
• scripts/agent_trace.sh (end-to-end agent smoke test with trace logs; set JCODE_PROVIDER=openai|claude)

MIT