AI Collaboration MCP Server

An MCP server for AI-to-AI collaboration, enabling autonomous workflows and role-based task management between different AI models.

🚧 Work in Progress - Active Development 🚧

A Model Context Protocol (MCP) server designed to facilitate direct AI-to-AI collaboration between Claude and Gemini, eliminating the need for human intermediation in development workflows.

Note: This project is under active development. While core features are functional, some aspects are still being refined. Contributions and feedback are welcome!

Enable truly autonomous AI-to-AI collaboration where:

- AI agents work continuously on complex projects
- Human intervention is minimal (ideally just starting the process)
- Agents create comprehensive project plans and execute 100+ phases autonomously
- Work continues until project completion or critical blocker

@ai-collab init {"agentName": "gemini", "autonomous": true} // For Gemini (CTO) @ai-collab init {"agentName": "claude", "autonomous": true} // For Claude (Developer)

- Loads existing project context and state
- Creates or resumes a comprehensive project plan
- Automatically detects and continues pending work
- Shows critical tickets and blockers
- Starts autonomous execution loops

- Task Dependencies: DefinedependsOnrelationships between tasks
- Batch Task Creation: CTO can create multiple tasks in one command
- Priority-Based Work: Tasks are automatically prioritized (high/medium/low)
- Continuous Developer Mode: No waiting between tasks - automatic progression
- Smart Task Status:available,blocked,in_progress,in_review,completed
- Dependency Resolution: Tasks automatically unblock when dependencies complete

- 120-second check intervalsfor more natural workflow pacing
- 500 iteration maximumfor extended autonomous operation
- Continuous work mode- agents keep working until project completion
- Manual loop execution- requires human to run check commands (automation WIP)

- Auto-generated 6-phase plansfrom PROJECT_REQUIREMENTS.md
- Smart phase progression- automatically moves to next phase when complete
- Duplicate task detection- prevents recreating completed features
- Ad-hoc mission support- pause main plan for urgent tasks

- Ticket vs Task distinction- prevents confusion between bug reports and work items
- Role-based instructions- clearer guidance for CTO vs Developer roles
- Workflow enforcement- ensures proper task creation and submission flow

- Manual loop execution required- AI agents can't schedule their own checks
- PATH configuration needed- Claude/Gemini commands must be accessible
- API quota limits- Gemini has daily request limits that may be exceeded

- Single task queuing→ Now supports batch task creation
- Developer idle time→ Continuous work mode implemented
- No task dependencies→ Full dependency system added
- Random task order→ Priority-based scheduling active

- Agents occasionally create duplicate tasks (improved but not eliminated)
- Edit button functionality may need manual verification
- Loop execution still requires human intervention

- Automation scripts provided (mcp-automator.js) but require setup
- Manual loop execution instructions included
- Simulation mode for tracking when automation fails

- Comprehensive Project Plans: 100+ phase autonomous execution capability
- One-Command Startup: Justinitwith autonomous flag
- Role-Based System: CTO, Developer, PM, QA, Architect roles
- Smart Task Management: Duplicate detection and phase progression
- Ticketing System: Track bugs, enhancements, tech debt
- Context Retention: Maintains state across sessions
- Mission Management: High-level objectives with auto-decomposition
- Code Review Workflow: Submit, review, and revision cycles
- Question & Answer System: Asynchronous clarifications
- Comprehensive Logging: Full audit trail

- Task Dependencies: Tasks can depend on other tasks with automatic blocking/unblocking
- Priority-Based Scheduling: High/medium/low priority with smart task selection
- Batch Task Creation: CTO can queue multiple tasks at once for efficiency
- Continuous Work Mode: Developer automatically moves to next available task
- Smart Status System:available,blocked,in_progress,in_review,completed
- Dependency Visualization: Clear indication of task dependencies and blockers

git clone https://github.com/yourusername/ai-collab-mcp.git cd ai-collab-mcp

Create a.mcp.jsonfile in your project root:

{ "mcpServers": { "ai-collab": { "command": "node", "args": [".mcp-server/src/index.js"], "cwd": "/path/to/your/project" } } }

{ "mcpServers": { "ai-collab": { "command": "node", "args": ["/path/to/ai-collab-mcp/src/index.js"] } } }

Note: Gemini may require explicit instructions to execute MCP commands.

Start with autonomous flag for continuous operation:

# Terminal 1 - Claude (Developer) @ai-collab init {"agentName": "claude", "autonomous": true} # Terminal 2 - Gemini (CTO) @ai-collab init {"agentName": "gemini", "autonomous": true} # Terminal 3 - Manual Loop Execution (Required) # Every 120 seconds, run: @ai-collab get_loop_status {"agentName": "claude"} @ai-collab get_loop_status {"agentName": "gemini"}

# Run automation script (requires setup) cd /path/to/project node mcp-automator.js auto # Or simulation mode (shows what would happen) node mcp-automator-v2.js auto

- send_directive- Create development tasks (now with dependencies & priority)
- send_batch_directives- Create multiple tasks at once
- review_work- Review submissions
- create_project_plan- Start comprehensive plan
- update_plan_progress- Move to next phase

- get_all_tasks- View assigned work (sorted by priority)
- submit_work- Submit completed tasks
- ask_question- Request clarification

// Single task with dependency @ai-collab send_directive { "taskId": "KAN-002", "title": "Create database tables", "specification": "Create user and project tables", "priority": "high", "dependsOn": ["KAN-001"] // Won't be available until KAN-001 is approved } // Batch creation with dependencies @ai-collab send_batch_directives { "tasks": [ { "taskId": "KAN-003", "title": "Setup database connection", "specification": "Configure PostgreSQL connection", "priority": "high" }, { "taskId": "KAN-004", "title": "Create user model", "specification": "Implement User model with validation", "priority": "medium", "dependsOn": ["KAN-003"] }, { "taskId": "KAN-005", "title": "Create auth endpoints", "specification": "Implement login/register endpoints", "priority": "medium", "dependsOn": ["KAN-004"] } ] }

When the developer runsget_loop_status, they will:
- See prioritized available tasks
- Automatically start on the highest priority task
- After submitting, immediately move to next task
- Continue until all available tasks are complete

No more waiting between tasks! The developer keeps working continuously.
- Automatic Plan Creation: On first init, generates 6-phase plan from requirements
- Phase Progression: Automatically advances when all phase tasks complete
- Duplicate Prevention: Skips tasks that match completed work
- Ad-hoc Missions: Can pause main plan for urgent work

- Foundation & Basic Structure
- Core Interactive Features
- UI/UX Enhancement
- Data Persistence
- Advanced Features
- Polish & Quality Assurance

data/ ├── tasks.json # Task tracking ├── missions.json # Active missions ├── project-state.json # Project configuration ├── project-plans.json # Comprehensive plans (NEW) ├── loop-states.json # Autonomous loop tracking (NEW) └── tickets/ └── tickets.json # Bug/enhancement tracking

- Prefix with: "Execute the following MCP command:"
- Or: "Use the ai-collab tool to run:"

- System now detects similar task names
- Manually clean duplicates fromdata/tasks.jsonif needed

- Ensure 120-second intervals between checks
- Verify agent hasn't exceeded maxIterations (500)
- Check API quotas haven't been exceeded

- True automation (removing manual loop execution)
- Better Gemini CLI integration
- Improved duplicate detection algorithms
- Cross-platform automation scripts
- Fork the repository
- Create feature branch (git checkout -b feature/improvement)
- Commit changes (git commit -m 'Add improvement')
- Push branch (git push origin feature/improvement)
- Open Pull Request

- Native scheduling in MCP server
- WebSocket/SSE for real-time updates
- Improved role switching
- Better error recovery
- Multi-project support
- Visual progress dashboard

MIT License - seeLICENSEfile for details.

A Python monorepo for AI-powered project management and productivity servers, utilizing the Claude API.