Agent Orchestrator MCP Server

Multi-agent orchestration for Claude with 65 tools. Turn Claude into a structured, multi-agent system with persistent memory, project scanning, standards enforcement, and formal task workflows.

Install

Claude Code

sudo npm install -g @tejalgnrml/agent-orchestrator-mcp
claude mcp add agent-orchestrator -s user -- agent-orchestrator-mcp

Verify:

claude mcp list

Claude Desktop

First install the package:

sudo npm install -g @tejalgnrml/agent-orchestrator-mcp

Then add to config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

{
  "mcpServers": {
    "agent-orchestrator": {
      "command": "agent-orchestrator-mcp"
    }
  }
}

Restart Claude Desktop after saving.

Update

sudo npm update -g @tejalgnrml/agent-orchestrator-mcp

Uninstall

claude mcp remove agent-orchestrator -s user
sudo npm uninstall -g @tejalgnrml/agent-orchestrator-mcp

What Is This?

Agent Orchestrator gives Claude structured multi-agent capabilities. Instead of Claude working as a single agent, it can:

• Create domain-specific agents (frontend, backend, API, etc.)
• Follow a structured workflow (Gather > Plan > Implement > Verify)
• Remember learnings across conversations
• Track project facts to prevent hallucination
• Enforce coding standards automatically
• Hand off work between agents with full context

Features (65 Tools)

| Feature | Tools | Description | |---------|-------|-------------| | Project Scanner | 3 | Auto-detect project structure and suggest agents | | Agent Memory | 3 | Persistent learnings that prevent repeated mistakes | | Project Standards | 4 | Define and enforce coding rules | | Project Facts | 4 | Store verified info (frameworks, versions, conventions) | | Agent Groups | 7 | Organize agents into teams | | Handoffs | 3 | Formal work transitions between agents | | Structured Tasks | 10 | Full GATHER > SYNTHESIZE > PLAN > IMPLEMENT > VERIFY workflow | | Audit Log | 2 | Track every action for debugging | | Agent Registry | 5 | Register agents with file ownership patterns | | Message Passing | 3 | Agent-to-agent communication | | Event System | 3 | Publish/subscribe events | | Proposals | 4 | Change approval workflows | | State Management | 2 | Persistent agent state | | Query System | 2 | Find agents by criteria or file ownership | | Coordination | 6 | Multi-agent session management | | Ownership Validation | 1 | Detect file ownership conflicts | | Workflow Guide | 1 | Get recommended workflows for any task type |

Use Cases

1. Auto-Setup Any Project

Tell Claude to scan your project and it will suggest agents based on your folder structure.

You: Initialize agents for this project
Claude: Scanning project...
        Detected: Next.js 14, TypeScript, PostgreSQL
        Suggested agents:
        - pages-agent (owns src/pages/**)
        - api-agent (owns src/api/**)
        - components-agent (owns src/components/**)
        - hooks-agent (owns src/hooks/**)
        Create these? [Yes/No]

2. Structured Feature Development

Claude follows a formal workflow instead of jumping straight to code.

You: Add user authentication with JWT

Claude follows this workflow:
  GATHER    - Each agent analyzes impact in their domain
  SYNTHESIZE - Combines all analyses into unified plan
  PLAN      - Creates ordered steps with dependencies
  IMPLEMENT - Executes steps, hands off between agents
  VERIFY    - Each agent validates changes in their domain

3. Prevent Hallucination with Facts

Store verified project information so Claude never guesses.

You: Store that we use Prisma ORM with PostgreSQL 15
Claude: [stores project fact]
        Stored! Will reference this in all future tasks.

Later...
You: Add a new database model
Claude: [queries facts first]
        I see you use Prisma with PostgreSQL 15. Creating model...

4. Remember Learnings Across Conversations

Claude remembers what worked and what didn't.

You: Remember that date formatting must use date-fns, not moment
Claude: [stores learning]
        Stored as convention. Will use date-fns in all future code.

5. Enforce Coding Standards

Define rules that Claude checks before and after writing code.

You: Set a standard that all API routes must have error handling
Claude: [sets standard: strict enforcement]
        Standard set! Will validate all API routes against this.

6. Multi-Agent Coordination

Different agents handle different parts of a large task.

You: Refactor the payment system

Claude:
  - api-agent handles backend routes
  - pages-agent handles UI changes
  - creates handoffs when work crosses domains
  - each agent verifies their area when done

7. Audit Trail

Track everything for debugging or review.

You: Show me what happened in the last task
Claude: [gets history]
        10:00 - Task started (auth feature)
        10:01 - api-agent gathered analysis
        10:02 - pages-agent gathered analysis
        10:03 - Synthesis completed
        10:05 - Step 1 executed (API routes)
        10:08 - Handoff from api-agent to pages-agent
        10:12 - Step 2 executed (Login page)
        10:15 - Verification passed

Workflows

Multi-Agent Task (Complex Features)

GATHER > SYNTHESIZE > PLAN > IMPLEMENT > VERIFY

Use for features that touch multiple parts of the codebase.

Single-Agent Task (Simple Changes)

ANALYZE > IMPLEMENT > VERIFY

Use for bug fixes or changes in one domain.

Project Setup

SCAN > REVIEW > APPLY

Use when setting up orchestration on a new project.

All 65 Tools

Project Scanner

| Tool | Description | |------|-------------| | initialize_project | Scan project and suggest agents/groups/facts | | apply_initialization | Create approved agents/groups/facts | | get_project_summary | View current orchestration setup |

Agent Registry

| Tool | Description | |------|-------------| | register_agent | Create agent with file ownership | | get_agent | Get agent details | | list_agents | List all agents | | update_agent | Modify agent config | | delete_agent | Remove agent |

Agent Memory

| Tool | Description | |------|-------------| | store_learning | Save knowledge (patterns, decisions, mistakes) | | get_learnings | Retrieve past learnings | | delete_learning | Remove outdated learning |

Project Standards

| Tool | Description | |------|-------------| | set_standard | Define coding rule with enforcement level | | get_standards | List all standards | | delete_standard | Remove standard | | validate_against_standards | Check files for violations |

Project Facts

| Tool | Description | |------|-------------| | store_project_fact | Record verified info (framework, DB, etc.) | | query_project_facts | Search stored facts | | update_fact | Modify a fact | | delete_fact | Remove a fact |

Agent Groups

| Tool | Description | |------|-------------| | create_group | Create agent team | | get_group | Get group details | | list_groups | List all groups | | update_group | Modify group | | delete_group | Remove group | | add_agent_to_group | Add member | | remove_agent_from_group | Remove member |

Handoffs

| Tool | Description | |------|-------------| | create_handoff | Transfer work between agents | | get_handoffs | List handoffs | | complete_handoff | Accept/reject/complete handoff |

Structured Tasks

| Tool | Description | |------|-------------| | start_task | Initialize multi-agent task | | gather_analysis | Agent provides analysis | | get_task_analyses | View all analyses | | store_synthesis | Combine analyses | | create_task_plan | Define implementation steps | | execute_step | Record step completion | | verify_changes | Agent verifies their domain | | complete_task | Mark task done/failed | | get_task | Get task details | | list_tasks | List all tasks |

Message Passing

| Tool | Description | |------|-------------| | send_message | Send message between agents | | get_messages | Get agent's messages | | mark_read | Mark messages as read |

Event System

| Tool | Description | |------|-------------| | publish_event | Publish event | | subscribe | Subscribe to event types | | get_events | Get events for agent |

Proposals

| Tool | Description | |------|-------------| | create_proposal | Propose a change for approval | | respond_to_proposal | Approve/reject | | get_proposals | List proposals | | update_proposal_status | Update status |

State Management

| Tool | Description | |------|-------------| | store_state | Save agent state | | get_state | Get agent state |

Query System

| Tool | Description | |------|-------------| | query_agents | Find agents by criteria | | find_agents_for_files | Find who owns specific files |

Coordination Sessions

| Tool | Description | |------|-------------| | analyze_task_scope | Determine which agents needed | | create_coordination_session | Start multi-agent session | | get_coordination_session | Get session details | | add_session_contribution | Agent contributes to session | | set_session_plan | Set coordinated plan | | update_session_status | Update session status | | list_coordination_sessions | List all sessions |

Ownership & Workflow

| Tool | Description | |------|-------------| | validate_ownership | Check for ownership conflicts | | get_workflow | Get recommended workflow | | get_history | View audit log | | get_agent_history | Agent-specific history |

Requirements

• Node.js >= 18.0.0
• Claude Code or Claude Desktop

License

PolyForm Noncommercial 1.0.0 - Free for personal and non-commercial use. See LICENSE for details.