claude-bedrock-cli 1.1.1

A powerful CLI for Claude AI using AWS Bedrock - no Anthropic subscription needed

Claude Bedrock CLI

A powerful command-line interface for Claude AI that uses AWS Bedrock instead of direct Anthropic API subscriptions. This is a full-featured implementation inspired by Claude Code, allowing you to interact with Claude, edit files, run commands, and manage complex tasks - all while using your existing AWS Bedrock subscription.

Features

• Full Claude Code Experience: Replicates core Claude Code functionality using AWS Bedrock
• File Operations: Read, write, and edit files with Claude's assistance
• Command Execution: Run bash/shell commands directly from the CLI
• Code Search: Search for patterns (grep) and find files (glob)
• Human-in-the-Loop: Approve tool executions before they run, with feedback capability
• Task Management: Built-in todo list for tracking complex tasks
• Rich Interface: Beautiful terminal UI with markdown rendering
• Conversation History: Persistent conversation across sessions
• Streaming Responses: Real-time streaming of Claude's responses

Prerequisites

• Python 3.8 or higher
• AWS Account with Bedrock access
• Claude model enabled in AWS Bedrock (e.g., Claude 3.5 Sonnet)
• AWS credentials configured

Installation

Option 1: Install from PyPI (Recommended for Users)

Simply install using pip:

pip install claude-bedrock-cli

That's it! The claude-bedrock command will be available globally.

Optional: Include AWS CLI

If you want the AWS CLI included (for advanced AWS management):

pip install claude-bedrock-cli[aws]

Note: The AWS CLI is optional - you can configure credentials using our built-in tool.

Option 2: Install from Source (For Developers)

If you want to modify the code:

# Clone the repository
git clone https://github.com/yourusername/claude-bedrock-cli.git
cd claude-bedrock-cli

# Install in development mode
pip install -e .

# Or with AWS CLI
pip install -e .[aws]

Configuration

1. Configure AWS credentials

You have three options for configuring AWS credentials:

Option A: Built-in Configuration Tool (Easiest - No AWS CLI Required)

claude-bedrock-configure

This interactive wizard will:

• Ask for your email/username (for usage tracking)
• Guide you through entering your AWS credentials
• Save them securely to ~/.aws/credentials
• Test your Bedrock connection
• Show available Claude models

Option B: AWS CLI (If Installed)

aws configure

Option C: Environment Variables

Create a .env file in your working directory:

AWS_ACCESS_KEY_ID=your_access_key_here
AWS_SECRET_ACCESS_KEY=your_secret_key_here
AWS_REGION=us-east-1
BEDROCK_MODEL_ID=anthropic.claude-3-haiku-20240307-v1:0

The CLI will automatically load these when you run it from that directory.

2. Enable Bedrock Model Access

Make sure you have access to Claude models in AWS Bedrock:

1. Go to AWS Console → Bedrock → Model access
2. Request access to Anthropic Claude models if not already enabled
3. Wait for approval (usually instant for Claude models)

Usage

Basic Usage

Run the CLI from any directory:

python -m claude_cli.main

Or if installed via setup.py:

claude-bedrock

Options

# Use a specific working directory
claude-bedrock --working-dir /path/to/project

# Use a different model
claude-bedrock --model-id anthropic.claude-3-opus-20240229-v1:0

# Continue from previous session (load conversation history)
claude-bedrock --continue

# Auto-approve all tool executions (skip confirmation prompts)
claude-bedrock --yes

# Combined
claude-bedrock -d /path/to/project -m anthropic.claude-3-haiku-20240307-v1:0 -c -y

Notes:

• By default, each session starts fresh. Use --continue or -c to load your previous conversation history.
• By default, you'll be prompted to approve tool executions (Write, Edit, Bash, etc.). Use --yes or -y to auto-approve all tools.

Available Commands

Startup Commands:

• claude-bedrock - Start the CLI
• claude-bedrock-configure - Configure AWS credentials (interactive wizard)

Interactive Commands (once in the CLI):

• Type your questions or requests naturally
• Type /clear to clear conversation history
• Type /todos to view the current todo list
• Type exit or quit to exit

Human-in-the-Loop Approval

By default, Claude will ask for your approval before executing tools like Write, Edit, or Bash commands. This gives you control over what changes are made to your system.

When a tool is about to execute, you'll see:

┌─ Tool Execution Request: Write ──────────┐
│ Write                                     │
│   File: login.html (new file)            │
│   Lines: 45                               │
│   Content preview:                        │
│                                           │
│ +   1 | <!DOCTYPE html>                  │
│ +   2 | <html>...                        │
└───────────────────────────────────────────┘

Approve this action? (Use ↑↓ arrows, press Enter)
> ✓ Approve - Execute this action
  ✗ Reject - Provide feedback
  ✓✓ Approve All - Auto-approve all remaining

Navigation:

• ↑/↓ Arrow Keys - Move between options
• Enter - Confirm selection
• Ctrl+C - Cancel

Options:

• ✓ Approve - Execute this tool
• ✗ Reject - Provide feedback on what to do differently
• ✓✓ Approve All - Auto-approve this and all remaining tools in the session

Example workflow:

> Create a login page

[Shows diff preview with line numbers]
Approve this action? (Use ↑↓ arrows, press Enter)
> ✓ Approve - Execute this action
  ✗ Reject - Provide feedback
  ✓✓ Approve All - Auto-approve all remaining

[Press ↓ to move to Reject]
  ✓ Approve - Execute this action
> ✗ Reject - Provide feedback
  ✓✓ Approve All - Auto-approve all remaining

[Press Enter]
✗ Rejected

Please provide feedback: Use a dark theme instead

[Claude adjusts and shows new diff]
[Press Enter on Approve]
✓ Approved
File created!

Example Interactions

File Operations:

> Read the README.md file and summarize it

> Create a new Python file called hello.py with a hello world function

> Find all Python files in this directory

Code Search:

> Search for the function definition of "calculate_total"

> Find all files containing "TODO"

> Show me all TypeScript files in the src directory

Command Execution:

> Run the tests using pytest

> Show me the git status

> Install the dependencies from requirements.txt

Complex Tasks:

> I need to refactor the authentication module to use JWT tokens instead of sessions

> Help me debug this error: [paste error]

> Create a REST API endpoint for user registration with proper validation

Available Tools

Claude has access to the following tools:

Read

Read files from the filesystem with optional line ranges.

Write

Create new files or overwrite existing ones.

Edit

Edit existing files by replacing exact string matches.

Bash

Execute shell commands and capture output.

Grep

Search for regex patterns in files with filtering options.

Glob

Find files matching glob patterns (e.g., **/*.py).

Project Structure

claude-bedrock-cli/
├── claude_cli/
│   ├── __init__.py
│   ├── main.py              # CLI entry point
│   ├── bedrock_client.py    # AWS Bedrock API wrapper
│   ├── tools.py             # Tool implementations
│   ├── conversation.py      # Conversation management
│   └── todo_manager.py      # Todo list management
├── requirements.txt
├── setup.py
├── .env.example
├── .gitignore
└── README.md

Configuration

Environment Variables

AWS Configuration

• AWS_ACCESS_KEY_ID: AWS access key (optional if using AWS CLI config)
• AWS_SECRET_ACCESS_KEY: AWS secret key (optional if using AWS CLI config)
• AWS_REGION: AWS region (default: us-east-1)
• BEDROCK_MODEL_ID: Bedrock model ID (default: anthropic.claude-3-haiku-20240307-v1:0)

LangSmith Logging (Optional - Admin Only)

• LANGSMITH_API_KEY: Your LangSmith API key for usage tracking
• LANGSMITH_PROJECT: Project name in LangSmith (default: claude-bedrock-cli)

Note: LangSmith logging is optional and tracks:

• User prompts and model responses
• Username/email for each interaction
• Model ID and tool usage
• Session information

If LangSmith credentials are not provided, logging is automatically disabled.

Available Models

Choose from these Bedrock model IDs:

• anthropic.claude-3-haiku-20240307-v1:0 - Latest Sonnet (recommended)
• anthropic.claude-3-opus-20240229-v1:0 - Most capable
• anthropic.claude-3-sonnet-20240229-v1:0 - Balanced
• anthropic.claude-3-haiku-20240307-v1:0 - Fastest

Cost Considerations

This CLI uses AWS Bedrock, which charges based on:

• Input tokens (what you send to Claude)
• Output tokens (what Claude generates)

Pricing varies by model and region. Check AWS Bedrock Pricing for current rates.

Tips to minimize costs:

• Use Claude 3.5 Sonnet for best balance of performance and cost
• Clear conversation history (/clear) for unrelated tasks
• Use Claude 3 Haiku for simple tasks

Troubleshooting

Getting Started Issues

Problem: Don't have AWS credentials set up

Solution: Use the built-in configuration tool:

claude-bedrock-configure

This will guide you through the setup process and test your connection.

"Access Denied" Error

Solution: Make sure you have:

1. Valid AWS credentials configured (run claude-bedrock-configure to verify)
2. Bedrock model access enabled in AWS Console
3. Proper IAM permissions for Bedrock

Required IAM permissions:

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Action": [
        "bedrock:InvokeModel",
        "bedrock:InvokeModelWithResponseStream"
      ],
      "Resource": "arn:aws:bedrock:*::foundation-model/anthropic.claude-*"
    }
  ]
}

"Model Not Found" Error

Solution:

1. Check that the model ID is correct
2. Verify model access is enabled in AWS Bedrock Console
3. Make sure you're using the correct region

Slow Response Times

Solution:

• Try a different AWS region closer to you
• Use a faster model (Claude 3 Haiku)
• Check your internet connection

Tool Execution Errors

Solution:

• Make sure you have proper file permissions
• Check that required commands (git, npm, etc.) are installed
• Verify the working directory is correct

Comparison with Claude Code

Feature	Claude Code	Claude Bedrock CLI
Claude API	Direct Anthropic	AWS Bedrock
File Operations	✅	✅
Command Execution	✅	✅
Code Search	✅	✅
Todo Management	✅	✅
Streaming	✅	✅
Multi-agent System	✅	⚠️ (Basic)
Plan Mode	✅	❌
Git Integration	✅	⚠️ (Via Bash)

Hosting Options

Local Usage

Simply run the CLI on your local machine.

Remote Server

You can deploy this on a remote server and SSH into it:

ssh your-server
cd ~/claude-bedrock-cli
claude-bedrock

Docker (Optional)

Create a Dockerfile:

FROM python:3.11-slim

WORKDIR /app
COPY . .

RUN pip install -e .

CMD ["claude-bedrock"]

Build and run:

docker build -t claude-bedrock-cli .
docker run -it --rm \
  -e AWS_ACCESS_KEY_ID=$AWS_ACCESS_KEY_ID \
  -e AWS_SECRET_ACCESS_KEY=$AWS_SECRET_ACCESS_KEY \
  -v $(pwd):/workspace \
  claude-bedrock-cli --working-dir /workspace

Contributing

This is a personal tool, but feel free to fork and modify it for your needs!

Security Notes

• Never commit your .env file or AWS credentials
• Use IAM roles when running on EC2/ECS instead of access keys
• Consider using AWS SSO for better security
• Limit IAM permissions to only what's needed

License

MIT License - feel free to use and modify as needed.

Support

For AWS Bedrock issues, refer to AWS Bedrock Documentation.

For Claude-specific questions, see Anthropic's Documentation.

Roadmap

Potential future enhancements:

• Multi-agent system (Plan, Explore agents)
• Plan mode for complex tasks
• Web search integration
• Image analysis support
• Custom slash commands
• Plugin system
• Configuration profiles
• Better error recovery

---

Enjoy using Claude with your AWS Bedrock subscription! 🚀