systems-thinking-mcp

An MCP server that provides systems thinking models as composable analysis lenses for architecture, infrastructure, DevOps, incident analysis, and technical decision-making.

Inspired by @modelcontextprotocol/server-sequential-thinking. Where sequential-thinking emphasizes rigor — structured step-by-step reasoning with revision and branching — this tool emphasizes abstraction. It provides a library of mental models (feedback loops, constraint analysis, inversion, etc.) that shift Claude's perspective on a problem. The two are complementary: use sequential-thinking to reason carefully through a problem, use systems-thinking to ensure Claude's looking at it from the right angles. This helps prevent tunnel vision, surface edge cases and unintended consequences, and reign in some of Claude's grandiousity by challenging the LLM to view the problem in the broader context in which it lives.

I've had good luck using this in troubleshooting and debugging, design, code review, and spec review (pairs nicely with superpower-mcp just run it against the produced spec after the spec review cylce). This is a highly general tool and can provide good results across different domains, development, infrastructure, devops.

How it works

Five tools with a lifecycle: start a session (returns model clusters by category), expand your selection (get full model details and graph neighbors), apply 2-4 lenses from different perspectives, then synthesize across them.

The value isn't in any single lens — it's in the composition. Each model surfaces things the others miss, and the server provides prior findings from earlier lenses so the LLM can judge connections between them. When you apply constraint analysis and then queuing theory, the findings from the first lens are available when applying the second.

Models also define counterbalances — deliberately opposing perspectives. When you apply leverage-points, the server suggests KISS as a counterbalance: "The simplest solution may miss high-leverage structural changes that pay off long-term." This productive tension prevents single-framework tunnel vision.

Related MCP server: Clear Thought MCP Server

Install

{
  "mcpServers": {
    "systems-thinking": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "systems-thinking-mcp"]
    }
  }
}

Or run locally:

git clone https://github.com/davidpanter/systems-thinking.git
cd systems-thinking
npm install && npm run build

{
  "mcpServers": {
    "systems-thinking": {
      "type": "stdio",
      "command": "node",
      "args": ["/path/to/systems-thinking/dist/index.js"]
    }
  }
}

Tools

Models (56)

Models support multi-facet categories via a categories array in YAML, allowing a single model to appear in multiple categories.

Strategies (8)

Strategies guide the LLM toward the right categories for a given task. Each strategy defines a concern map — a list of domains (matching category names) with a focus question and weight (required, conditional, optional). Strategies work for any system — code, infrastructure, pipelines, platforms, or design documents. Strategy-to-model validation runs at startup, ensuring concern domains match actual category names.

Custom models

Add your own models via --models-dir. Custom models with the same ID as built-in models override them.

{
  "mcpServers": {
    "systems-thinking": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "systems-thinking-mcp", "--models-dir", "/path/to/models"]
    }
  }
}

Models are YAML files in category subdirectories. See models/ for the format.

Environment variables

• DISABLE_THOUGHT_LOGGING=true — suppress stderr logging