MCP Server Setup

The MCP (Model Context Protocol) server provides the deepest integration with Claude Code, exposing AI Mantras as native tools.

What You Get

Core Tools

Tool	Description
`bootstrap_session`	Initialize a session with all needed resources
`assess_complexity`	Automatically triage requests
`get_persona`	Load personas by name or domain
`get_pattern`	Load thinking patterns
`get_skill`	Load skills by name, task, or category
`get_workflow`	Get orchestration workflows
`create_handoff`	Generate handoff templates
`list_available`	Discover all personas, patterns, and skills

Multi-Agent Tools (when enabled)

Tool	Description
`spawn_agent`	Spawn isolated agents with specific personas
`get_agent_result`	Retrieve results from spawned agents
`list_agents`	Monitor all spawned agents

Install the package globally:
Terminal window
```
npm install -g @ai-mantras/mcp-server
```

Add to your Claude Code MCP configuration (~/.claude/mcp.json or project .mcp.json):

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

Clone and build:

git clone https://github.com/kd5ziy/AIMantras.git
cd AIMantras/ai-mantras-mcp
npm install
npm run build

Add to your MCP configuration:

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

Multi-Agent Mode (Optional)

By default, AI Mantras runs in single-agent mode where all personas share one AI model. To enable multi-agent mode where each persona spawns as an isolated agent:

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

Multi-Agent Benefits

True context isolation between personas
Each agent only sees its persona, task, and inputs
Support for parallel execution
Mix Anthropic and OpenAI models per persona

Environment Variables

Variable	Description	Default
`MANTRAS_MULTI_AGENT_ENABLED`	Enable multi-agent tools	`false`
`ANTHROPIC_API_KEY`	API key for Claude models	-
`OPENAI_API_KEY`	API key for GPT models	-
`MANTRAS_DEFAULT_PROVIDER`	Default: `anthropic` or `openai`	`anthropic`
`MANTRAS_MAX_CONCURRENT_AGENTS`	Max parallel agents	`5`
`MANTRAS_AGENT_TIMEOUT_MS`	Agent timeout (ms)	`120000`

Usage

Once configured, Claude Code automatically has access to AI Mantras tools. You can:

Ask questions and let the framework triage automatically
Request specific personas: “Use Clara to analyze this investment”
Apply patterns: “Use chain-of-thought to work through this problem”