Claude Code

Overview

Claude Code is Anthropic’s CLI coding agent. It supports MCP servers, allowing AgentZero to provide policy-enforced tools (read, write, search, edit, shell) to Claude Code sessions.

In this mode, Claude Code drives the LLM (Claude); AgentZero provides tools with security policies, path validation, and audit logging.

Setup

1. Build with MCP support

cargo build --release --features mcp

2. Add to Claude Code MCP config

Create or edit .mcp.json in your project root:

{
  "mcpServers": {
    "agentzero": {
      "command": "az",
      "args": ["mcp"]
    }
  }
}

Or use the automated setup:

az connect claude --dry-run  # preview the config
az connect claude             # write to .mcp.json

3. Verify

Start Claude Code in your project directory. AgentZero tools will appear alongside Claude Code’s built-in tools.

Available Tools

Tool	Description
read_file	Read file contents (policy-checked)
list_directory	List directory contents
search_files	Search for patterns in files
write_file	Write content to a file (requires approval)
run_command	Execute a shell command (requires approval)

Security

• All tool calls go through AgentZero’s policy engine
• Path validation blocks access outside project root
• Sensitive paths (.ssh, .env, .aws/credentials) are denied
• Audit events emitted for every call to .agentzero/audit/
• Project policy loaded from .agentzero/policy.yml

Troubleshooting

• Tools not visible — ensure az is in your PATH and built with --features mcp
• Policy denials — check .agentzero/policy.yml or run az policy status
• Audit log — inspect with az audit tail