AI Mantras MCP Server

An MCP (Model Context Protocol) server that exposes the AI Mantras prompt framework as structured tools, resources, and prompts for Claude.

What is AI Mantras?

AI Mantras is a prompt engineering framework that treats AI collaboration as organizational design. It provides:

• 11 Personas across orchestration, domain, and evaluation categories
• 9 Thinking Patterns for structured reasoning
• 24 Skills - actionable capabilities personas can invoke
• 3 Complexity Tiers (Simple, Moderate, Complex) with appropriate workflows
• Separation of Powers to prevent self-evaluation and ensure quality

Installation

For Users (Zero Config)

# Install globally
npm install -g @ai-mantras/mcp-server

# Or use with npx (no install needed)

Add to your Claude Code .mcp.json:

{
  "mcpServers": {
    "ai-mantras": {
      "command": "npx",
      "args": ["-y", "@ai-mantras/mcp-server"]
    }
  }
}

For Development

cd ai-mantras-mcp
npm install
npm run build

Tools

Core Tools

| Tool | Description | |------|-------------| | bootstrap_session | Initialize a session with appropriate resources for a request | | assess_complexity | Triage a request into Simple/Moderate/Complex tier | | get_persona | Load a persona by name or domain match | | get_pattern | Load a thinking pattern by name | | get_skill | Load a skill by name, task match, or category | | get_workflow | Get workflow steps for a complexity tier | | create_handoff | Generate handoff template between personas | | list_available | List all available personas, patterns, and skills |

Multi-Agent Tools

These tools enable true multi-agent execution with context isolation between personas.

Note: These tools only appear when MANTRAS_MULTI_AGENT_ENABLED=true

| Tool | Description | |------|-------------| | spawn_agent | Spawn an isolated AI agent with a specific persona | | get_agent_result | Retrieve status and result from a spawned agent | | list_agents | List all spawned agents with status and usage stats |

spawn_agent

Spawns an isolated agent that only sees its persona definition, the task, and any provided inputs. Supports both synchronous (wait for result) and asynchronous (return immediately) modes.

spawn_agent({
  persona: "Clara-Financial-Analyst",  // Required: persona name
  task: "Analyze this investment opportunity", // Required: what to do
  inputs: { document: "..." },          // Optional: context data
  success_criteria: "Provide ROI estimate", // Optional: success definition
  patterns: ["chain-of-thought"],       // Optional: thinking patterns
  provider: "anthropic",                // Optional: "anthropic" or "openai"
  model: "claude-sonnet-4-20250514",   // Optional: specific model
  async: false,                         // Optional: async mode
  max_tokens: 4096,                     // Optional: response limit
  timeout_ms: 120000                    // Optional: timeout
})

get_agent_result

Poll for completion or retrieve results from a spawned agent.

get_agent_result({
  agent_id: "agent_abc12345"  // Required: ID from spawn_agent
})

list_agents

List all spawned agents with optional status filtering.

list_agents({
  status: "all",  // Optional: pending|running|completed|failed|timeout|all
  limit: 20       // Optional: max results
})

Resources

| URI | Content | |-----|---------| | mantras://principles | Guiding principles | | mantras://bootstrapper | Agent bootstrapper | | mantras://manifest | Framework manifest (YAML) | | mantras://toolset | Skills master index | | mantras://persona/{category}/{name} | Individual personas | | mantras://pattern/{layer}/{name} | Individual patterns | | mantras://skill/{category}/{name} | Individual skills |

Prompts

| Prompt | Description | |--------|-------------| | mantras | Auto-triage and process request | | mantras-simple | Direct answer with single persona | | mantras-moderate | Structured reasoning + self-review | | mantras-complex | Full orchestration workflow |

Usage Examples

Using the main prompt

/mantras "What's the best filesystem for a NAS?"

Using tools directly

Use the assess_complexity tool to analyze: "Help me design a backup strategy"

Reading resources

Read mantras://persona/domain/Kestra-Systems-Architect

Environment Variables

Operating Modes

AI Mantras supports two operating modes:

| Mode | Description | Use Case | |------|-------------|----------| | Single-Agent (default) | All personas run on one AI model | Simple setup, no API keys needed, lower cost | | Multi-Agent | Each persona spawns as a separate isolated agent | True context isolation, parallel execution, model specialization |

Set the mode via environment variable:

MANTRAS_MULTI_AGENT_ENABLED=false  # Single-agent mode (default)
MANTRAS_MULTI_AGENT_ENABLED=true   # Multi-agent mode

Multi-Agent Configuration

When MANTRAS_MULTI_AGENT_ENABLED=true, configure these additional settings:

| Variable | Description | Required | |----------|-------------|----------| | MANTRAS_MULTI_AGENT_ENABLED | Enable multi-agent tools | No (default: false) | | ANTHROPIC_API_KEY | Anthropic API key for Claude models | For Anthropic provider | | OPENAI_API_KEY | OpenAI API key for GPT models | For OpenAI provider | | MANTRAS_DEFAULT_PROVIDER | Default provider: anthropic or openai | No (default: anthropic) | | MANTRAS_DEFAULT_MODEL | Default model name | No (auto-selected) | | MANTRAS_MAX_CONCURRENT_AGENTS | Max parallel async agents | No (default: 5) | | MANTRAS_AGENT_TIMEOUT_MS | Default timeout in ms | No (default: 120000) |

Example Configurations

Single-Agent Mode (default, no extra config needed):

{
  "mcpServers": {
    "ai-mantras": {
      "command": "npx",
      "args": ["-y", "@ai-mantras/mcp-server"]
    }
  }
}

Multi-Agent Mode (enables spawn_agent, get_agent_result, list_agents):

{
  "mcpServers": {
    "ai-mantras": {
      "command": "npx",
      "args": ["-y", "@ai-mantras/mcp-server"],
      "env": {
        "MANTRAS_MULTI_AGENT_ENABLED": "true",
        "ANTHROPIC_API_KEY": "sk-ant-..."
      }
    }
  }
}

Custom Content

By default, the server uses bundled content. Power users can override with custom content:

{
  "mcpServers": {
    "ai-mantras": {
      "command": "npx",
      "args": ["-y", "@ai-mantras/mcp-server"],
      "env": {
        "MANTRAS_CONTENT_PATH": "/path/to/your/content"
      }
    }
  }
}

Available Personas

Orchestration

• Bernstein-Orchestrator - Coordinates multi-persona workflows
• Hopper-Project-Planner - Frames problems and decomposes work
• Lovell-Crisis-Planner - Handles high-stakes, time-sensitive planning

Domain

• Clara-Financial-Analyst - Financial analysis and portfolio strategy
• Kestra-Systems-Architect - Infrastructure and systems design
• Goeth-Philosophical-Synthesizer - Ethics and meaning frameworks
• Franklin-Deep-Reasoner - Multi-domain integration
• Watson-Medical-Advisor - Evidence-based medical guidance

Evaluation

• Ada-QA-Reviewer - Quality assurance and review
• Drucker-Goal-Satisfaction-Evaluator - Objective measurement
• Rickover-Safety-Evaluator - Risk and safety assessment

Available Patterns

Layer 1: Foundational

• planning-phase - Problem definition before execution
• orchestration - Multi-persona coordination
• recursive-self-eval - Critique and refinement loops
• meta-rules - Behavioral guardrails

Layer 2: Thinking Primitives

• chain-of-thought - Labeled reasoning stages
• rule-based-reasoning - Explicit rules with traceability
• guardrail-creative - Creativity within constraints

Layer 3: Evaluation

• criterion-based-evaluation - Systematic assessment
• threat-modeling - Risk identification

Available Skills

Research

• web-search - Search and synthesize web information
• codebase-exploration - Navigate and understand code

Analysis

• code-review - Review code for quality and issues
• risk-assessment - Evaluate risks and mitigations

Creation

• document-generation - Create structured documents
• code-generation - Generate code from specifications

Evaluation

• quality-check - Verify output quality
• security-audit - Assess security posture

Orchestration

• task-decomposition - Break down complex tasks
• handoff - Transfer work between personas

Utility

• help - Framework guidance and discovery

License

MPL-2.0