Skip to content

claude-code.md

Latest commit

 

History

History
144 lines (109 loc) · 3.88 KB

claude-code.md

File metadata and controls

144 lines (109 loc) · 3.88 KB

Claude Code Provider

Visor includes advanced integration with the Claude Code SDK, providing powerful AI-driven code analysis with MCP (Model Context Protocol) tools and subagent support.

Features

  • Advanced AI Analysis: Leverages Claude Code's sophisticated understanding
  • MCP Tools: Custom tools for specialized analysis via MCP servers
  • Subagents: Delegate specific tasks to specialized agents
  • Session Reuse: Continue conversations across dependent checks
  • Flexible Permissions: Granular control over tool usage

Configuration Options

The claude_code configuration block supports the following options:

Option Type Description
allowedTools string[] List of tool names Claude can use (e.g., ['Grep', 'Read', 'WebSearch'])
maxTurns number Maximum conversation turns (default: 5)
systemPrompt string Custom system prompt for the analysis
mcpServers object MCP server configurations (see below)
subagent string Path to a subagent definition file
hooks object Lifecycle hooks (onStart, onEnd, onError)

Basic Example

steps:
  claude_security_review:
    type: claude-code
    prompt: "Perform a comprehensive security review"
    claude_code:
      allowedTools: ['Grep', 'Read', 'WebSearch']
      maxTurns: 5
      systemPrompt: "You are an expert security auditor"

Example with MCP Servers

steps:
  claude_with_mcp:
    type: claude-code
    prompt: "Analyze code complexity and architecture"
    claude_code:
      allowedTools: ['Read', 'Grep', 'custom_tool']
      mcpServers:
        custom_analyzer:
          command: "node"
          args: ["./mcp-servers/analyzer.js"]
          env:
            ANALYSIS_MODE: "deep"

MCP Server Configuration

Each MCP server entry supports:

Option Type Required Description
command string Yes The command to spawn the MCP server
args string[] No Command line arguments
env object No Environment variables for the server process

Subagents

Use the subagent option to delegate tasks to specialized agents:

steps:
  claude_comprehensive:
    type: claude-code
    prompt: "Perform a comprehensive code review"
    claude_code:
      subagent: "./.claude/agents/code-reviewer.md"
      maxTurns: 10

Hooks

Lifecycle hooks allow custom processing at different stages:

steps:
  claude_with_hooks:
    type: claude-code
    prompt: "Review the code"
    claude_code:
      hooks:
        onStart: "echo 'Starting review'"
        onEnd: "echo 'Review complete'"
        onError: "./scripts/handle-error.sh"

Session Reuse

Reuse Claude Code sessions across dependent checks for context continuity:

steps:
  initial_analysis:
    type: claude-code
    prompt: "Analyze the code structure"
    claude_code:
      maxTurns: 3

  follow_up:
    type: claude-code
    prompt: "Based on the previous analysis, suggest improvements"
    depends_on: [initial_analysis]
    reuse_ai_session: true

Environment Variables

The provider requires an API key via one of these environment variables:

# Option 1: Claude Code specific key
export CLAUDE_CODE_API_KEY=your-api-key

# Option 2: General Anthropic API key
export ANTHROPIC_API_KEY=your-api-key

Required Dependencies

Install the Claude Code SDK:

npm install @anthropic/claude-code-sdk @modelcontextprotocol/sdk

Complete Example

See examples/claude-code-config.yaml for a comprehensive configuration example.

Related Documentation