Claude Code

Anthropic's official AI coding agent with extended thinking, plugins, and subagents.

Key Features

Extended Thinking

See Claude's reasoning process in real-time. Watch as it analyzes code, considers options, and explains its approach before taking action.

Plugins & Skills

Extend capabilities with plugins. Load specialized skills for frameworks, languages, or custom workflows.

Subagents

Spawn specialized agents for focused tasks. Create a code reviewer, documentation writer, or test generator that runs in parallel.

File Checkpointing

Every user message creates a checkpoint. Undo file changes to any point in the conversation with a single action.

Setup

Installation

bash

# Install Claude Code CLI
curl -fsSL https://claude.ai/install.sh | bash

# Verify installation
claude --version

Authentication

Choose one authentication method:

Option A: API Key (Recommended)

Set your Anthropic API key as an environment variable. Best for automated setups.

bash

# Add to your shell profile (~/.zshrc or ~/.bashrc)
export ANTHROPIC_API_KEY=sk-ant-your-key-here

# Then start the bridge
broski

Option B: OAuth Login

Use browser-based authentication. Convenient for personal use.

bash

# Authenticate via browser
claude login

# Then start the bridge
broski

Permission Modes

Claude Code has different permission modes that control how it handles file edits and shell commands:

Mode	File Edits	Shell Commands	Use Case
`default`	Ask	Ask	Safe exploration, review each change
`acceptEdits`	Auto-approve	Ask	Trust file changes, review commands
`plan`	Blocked	Blocked	Read-only analysis, no modifications
`bypassPermissions`	Auto-approve	Auto-approve	Full trust, automated workflows

Use bypassPermissions only in trusted environments. The agent can execute any command without confirmation.

Built-in Tools

Tool	Description	Permission Required
`Read`	Read file contents with optional line range	Safe
`Write`	Create new files (requires prior read)	Approval
`Edit`	String replacement in existing files	Approval
`Bash`	Execute shell commands with timeout	Approval
`Glob`	Find files matching pattern (e.g., */.ts)	Safe
`Grep`	Regex search in file contents	Safe
`WebFetch`	Fetch web page contents	Safe
`WebSearch`	Search the web for documentation	Safe
`Task`	Spawn subagents for parallel work	Safe
`TodoRead`	Read current task list	Safe
`TodoWrite`	Update task list	Safe

Subagents

Define specialized agents for focused tasks in your project's .claude/ directory:

.claude/settings.json

{
  "agents": {
    "code-reviewer": {
      "description": "Expert code review focusing on security and best practices",
      "prompt": "Review code for security vulnerabilities, performance issues, and best practices",
      "tools": ["Read", "Grep", "Glob"],
      "model": "sonnet"
    },
    "test-writer": {
      "description": "Generate comprehensive test cases",
      "prompt": "Write thorough unit and integration tests",
      "tools": ["Read", "Write", "Glob"],
      "model": "sonnet"
    }
  }
}

MCP Configuration

Add MCP servers to extend Claude Code's capabilities:

~/.claude.json

{
  "mcpServers": {
    "filesystem": {
      "type": "stdio",
      "command": "npx",
      "args": ["@modelcontextprotocol/server-filesystem"],
      "env": {
        "ALLOWED_PATHS": "/Users/me/projects"
      }
    },
    "github": {
      "type": "stdio",
      "command": "npx",
      "args": ["@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_TOKEN": "ghp_your_token_here"
      }
    }
  }
}

MCP tools appear with a prefix like mcp__filesystem__list_files. You can allow or block specific MCP tools in your settings.

Session Storage

Claude Code stores sessions as JSONL files on your computer:

~/.claude/projects/[project-path-encoded]/[session-id].jsonl

Feature	Description
Resume	Continue any previous conversation
Fork	Branch from any message (creates new file)
Delete	Remove session files to free space
Transcript	Full history with tool calls and results

Troubleshooting

Authentication Failures

Auth not configured

No valid authentication found. Ensure ANTHROPIC_API_KEY is set or run claude login.

bash

# Check if API key is set
echo $ANTHROPIC_API_KEY

# If empty, add to your shell profile:
echo 'export ANTHROPIC_API_KEY=sk-ant-your-key' >> ~/.zshrc
source ~/.zshrc

# Or re-authenticate:
claude logout
claude login

Rate Limiting

Rate limit exceeded

Too many requests. The bridge will automatically retry with backoff.

Check your usage at the Anthropic Console. Higher tier plans have increased rate limits.

Context Length

Context too long

Conversation exceeds model context window.

Start a new session or use the summarize feature to compact the conversation. Long sessions with many file operations can hit context limits.

Tool Permission Denied

Tool not allowed

The requested tool is disabled in your configuration.

~/.claude/settings.json

{
  "allowedTools": ["Read", "Write", "Edit", "Bash", "Glob", "Grep"],
  "disallowedTools": []
}

OpenCode

Multi-provider alternative

Backend Overview

Compare all backends

Troubleshooting

More backend issues

Claude Code

Key Features

Setup

Installation

Authentication

Permission Modes

Built-in Tools

Subagents

MCP Configuration

Session Storage

Troubleshooting

Authentication Failures

Rate Limiting

Context Length

Tool Permission Denied

Next