claude-command 0.6.0

Claude Command - AI command center for Claude Code

Claude Command MCP Server

Real-time AI command center with unified naming, session indexing, and multi-provider orchestration.

[!IMPORTANT] System Prompt: Development standards and procedures are documented in .air/prompts/system.prompt.md - this is the most important file for understanding how to work with this codebase.

Overview

Claude Command is an MCP (Model Context Protocol) server that enables sophisticated AI mission orchestration across multiple providers. Built with fastMCP, it features a unified naming system, JSONL session indexing, and segregated conversation management for different mission types.

Features

• Mission-Based Architecture: Specialized AI interactions (brainstorm, recon, query, review)
• Multi-Provider Support: Gemini, OpenAI, and Anthropic models
• Unified Naming System: {mission-type}-{subject-slug}-{timestamp} pattern
• JSONL Session Index: Efficient append-only session tracking
• Segregated Conversations: Provider isolation for reconnaissance missions
• Search & Discovery: Full-text search across all sessions
• Real-Time Streaming: Watch conversations unfold live via tail -f
• Session Management: Persistent conversation storage and retrieval

Architecture

Built using fastMCP framework with modular mission system and unified session management:

Claude Code CLI ↔ fastMCP Server ↔ Mission System ↔ [Gemini ↔ OpenAI ↔ Anthropic]
                                        ↓
                        [Query | Review | Brainstorm | Recon]
                                        ↓
                      [Unified Naming | JSONL Index | Session Management]

Mission Types

• Brainstorm: Shared conversation history (group forum style)
• Recon: Private conversation history per provider (segregated intelligence)
• Review: Iterative conversation history per provider (code feedback)
• Query: Stateless execution (no conversation history)

Installation

Prerequisites

• Claude Code CLI
• Python 3.10+
• API keys for desired providers

Install Dependencies

git clone https://github.com/shaneholloman/mcp-claude-command
cd mcp-claude-command
uv sync

Configure API Keys

Set environment variables:

export GEMINI_API_KEY="your_gemini_key"
export OPENAI_API_KEY="your_openai_key"
export ANTHROPIC_API_KEY="your_anthropic_key"

Add to Claude Code

Install as a global UV tool and configure with Claude CLI:

# Install claude-command globally
uv tool install .

# Configure with Claude CLI
claude mcp add claude-command claude-command \
  -e GEMINI_API_KEY="$GEMINI_API_KEY" \
  -e CLAUDE_COMMAND_OPENAI_API_KEY="$OPENAI_API_KEY" \
  -e CLAUDE_COMMAND_ANTHROPIC_API_KEY="$ANTHROPIC_API_KEY" \
  --scope user

# Verify configuration
claude mcp list
claude mcp get claude-command

[!IMPORTANT] Restart Claude's CLI after configuration.

Usage

Available Tools

Tool	Description	Key Parameters
claude_command_query	Multi-provider query or conversation	mission_prompt, providers, temperature
claude_command_brainstorm	Creative collaboration with shared context	topic, context, providers, temperature
claude_command_recon	Independent reconnaissance with segregated conversations	mission_prompt, providers, temperature
claude_command_review	Code review with iterative feedback	code, focus, providers, temperature
claude_command_missions_list	List mission runs with optional filtering	mission_type
claude_command_missions_search	Search mission runs by content	query
claude_command_missions_rebuild_index	Rebuild mission index from files	-
claude_command_status	Server health and configuration	-
claude_command_conversations_list	List conversation files	-

Mission Examples

Brainstorm Mission (Shared History)

# All providers collaborate with shared conversation context
claude_command_brainstorm topic="API testing strategies" context="Microservices architecture" providers="gemini,openai,anthropic"

Result Files:

• brainstorms/brainstorm-api-testing-strategies-2025-06-16-14-30-22.json
• conversations/brainstorm-api-testing-strategies.jsonl (shared context)
• Stream files: streams/brainstorm-api-testing-strategies-2025-06-16-14-30-22-{provider}.md

Recon Mission (Segregated History)

# Each provider maintains private conversation with Claude
claude_command_recon mission_prompt="What are the latest security vulnerabilities?" providers="gemini,openai,anthropic"

Result Files:

• recons/recon-latest-security-vulnerabilities-2025-06-16-14-30-22.json
• missions/mission-recon-2025-06-16-14-30-22/ containing:
  • gemini-context.jsonl (private conversation)
  • openai-context.jsonl (private conversation)
  • anthropic-context.jsonl (private conversation)

Query Mission (Stateless)

# Independent responses with no conversation history
claude_command_query mission_prompt="Explain GraphQL vs REST" providers="gemini,openai,anthropic"

Mission Management

List Mission Runs

# List all mission runs
claude_command_missions_list

# Filter by mission type
claude_command_missions_list mission_type="recon"

Search Mission Runs

# Search by content
claude_command_missions_search query="API security"
claude_command_missions_search query="machine learning"

Rebuild Index

# Rebuild mission index from existing files
claude_command_missions_rebuild_index

File Organization

Unified Naming Pattern

All mission files follow the pattern: {mission-type}-{subject-slug}-{timestamp}.{extension}

Examples:

• brainstorm-api-testing-strategies-2025-06-16-14-30-22.json
• recon-security-vulnerabilities-2025-06-16-14-30-22.json
• query-graphql-vs-rest-comparison-2025-06-16-14-30-22.json

Directory Structure

~/claude-command/
├── index.jsonl                  # JSONL mission index (searchable)
├── brainstorms/                 # Brainstorm mission results
│   └── brainstorm-{subject-slug}-{timestamp}.json
├── recons/                      # Recon mission results
│   └── recon-{subject-slug}-{timestamp}.json
├── queries/                     # Query mission results
│   └── query-{subject-slug}-{timestamp}.json
├── reviews/                     # Review mission results
│   └── review-{subject-slug}-{timestamp}.json
├── conversations/               # Shared conversation files (JSONL)
│   └── brainstorm-{subject-slug}.jsonl
├── missions/                    # Segregated conversation sessions
│   └── mission-recon-{timestamp}/
│       ├── gemini-context.jsonl
│       ├── openai-context.jsonl
│       └── anthropic-context.jsonl
├── streams/                     # Real-time streaming files
│   ├── {mission}-{subject}-{timestamp}-{provider}.md
│   └── {mission}-{subject}-{timestamp}-summary.md
└── histories/                   # Complete provider histories
    ├── recon-{provider}-complete.jsonl
    └── review-{provider}-complete.jsonl

File Types

• Mission Results: JSON files with structured mission outcomes
• Conversations: JSONL files for shared conversation history
• Session Contexts: JSONL files for provider-specific conversations
• Stream Files: Markdown files with real-time conversation streams
• Mission Index: JSONL file with searchable mission run metadata

Mission Index System

JSONL Mission Index

The index.jsonl file contains one JSON object per line, enabling efficient append operations:

{"id":"recon-api-security-2025-06-16-14-30-22","mission_type":"recon","subject":"API Security","subject_slug":"api-security","timestamp":"2025-06-16-14-30-22","mission_prompt":"What are API security best practices?","providers":["gemini","openai","anthropic"],"result_files":{"main":"/path/to/result.json","summary":"/path/to/summary.md"},"created_at":"2025-06-16T14:30:22.123456"}

Mission Discovery

• List by Type: Filter mission runs by mission type (brainstorm, recon, query, review)
• Full-Text Search: Search across subjects, prompts, and mission IDs
• Chronological: Mission runs automatically sorted by timestamp (newest first)
• Rebuild: Reconstruct index from existing files when needed

Mission Architecture

Conversation History Patterns

Brainstorm (Shared History)

• All providers see shared conversation in conversations/brainstorm-{subject}.jsonl
• Group forum dynamics with cross-provider awareness
• Each round builds on all previous responses

Recon (Segregated History)

• Each provider maintains private conversation in missions/mission-recon-{timestamp}/{provider}-context.jsonl
• Complete provider isolation prevents cross-contamination
• Independent intelligence gathering for unbiased comparison

Review (Iterative History)

• Provider-specific conversation history in missions/mission-review-{timestamp}/{provider}-context.jsonl
• Iterative code review conversations with context building
• Each provider builds on their own review history

Query (Stateless)

• No conversation history maintained
• Each execution is completely independent
• Pure request-response pattern

Real-Time Streaming

Watch live conversations with tail -f:

# Watch specific provider
tail -f ~/claude-command/streams/recon-api-security-2025-06-16-14-30-22-gemini.md

# Watch summary
tail -f ~/claude-command/streams/recon-api-security-2025-06-16-14-30-22-summary.md

# Multi-pane viewing (tmux commands provided in summary files)

Configuration

Environment Variables

• GEMINI_API_KEY - Gemini API key
• OPENAI_API_KEY - OpenAI API key
• ANTHROPIC_API_KEY - Anthropic API key
• CONVERSATIONS_DIR - Base directory (default: ~/claude-command)
• DEFAULT_TEMPERATURE - Default temperature (default: 0.7)

Default Models

• Gemini: gemini-2.0-flash-exp
• OpenAI: gpt-4
• Anthropic: claude-3-5-sonnet-20241022

Development

Project Structure

src/claude_command/
├── server.py                    # MCP server with session management tools
├── settings.py                  # Configuration management
├── missions/                    # Mission implementations
│   ├── brainstorm.py           # Shared conversation missions
│   ├── recon.py                # Segregated conversation missions
│   ├── query.py                # Stateless query missions
│   └── review.py               # Iterative review missions
├── storage/                     # Storage and session management
│   ├── manager.py              # Session index and file operations
│   └── naming.py               # Unified naming utilities
├── utils/                       # General utilities
│   └── text.py                 # Text processing and slug generation
├── streaming.py                 # Real-time streaming infrastructure
├── dialog/                      # Conversation management
└── clients/                     # AI provider implementations

Local Development

git clone https://github.com/shaneholloman/mcp-claude-command
cd mcp-claude-command
uv sync
uv run python -m claude_command --gemini-api-key YOUR_KEY

Testing

Run comprehensive mission testing:

# Test all mission types with all providers
python3 -c "
from src.claude_command.missions.brainstorm import execute_brainstorm
from src.claude_command.missions.recon import execute_recon
from src.claude_command.missions.query import execute_query

# Test brainstorm (shared history)
execute_brainstorm('AI testing strategies', 'Validation approach', 'gemini,openai,anthropic', 0.3, True)

# Test recon (segregated history)
execute_recon('API security patterns', 'gemini,openai,anthropic', 0.3, True)

# Test query (stateless)
execute_query('GraphQL vs REST', 'gemini,openai,anthropic', 0.3, True)
"

Migration Ready

This system is designed for seamless migration to React TypeScript + Convex:

• API-Style Naming: Predictable patterns for database schema
• JSONL Format: Efficient append operations map to database inserts
• Mission Index: Ready for real-time search with Convex snapshots
• Structured Data: JSON mission results translate directly to database records

Requirements

• Claude Code CLI (will NOT work with Claude Desktop)
• At least one API key (Gemini, OpenAI, or Anthropic)
• Python 3.10+
• uv package manager

License

MIT License - see LICENSE for details.