PromptSmith

Reasoning as a Service - A Capability-Level 0 MCP Server that provides intelligent reasoning framework selection and meta-prompt generation for AI agents.

Purpose

PromptSmith analyzes tasks, determines their complexity and category, and generates optimized "meta-prompts" using the most appropriate reasoning framework.

Features

• 🧠 40 Reasoning Frameworks: Chain of Thought, Tree of Thoughts, ReAct, Reflexion, Graph of Thoughts, Program of Thoughts, and more
• 📊 Intelligent Task Analysis: Automatic category detection (code, math, logic, creative, research, data, planning) and complexity scoring (0–10)
• 🎯 Framework Selection: Heuristic-based selection of the optimal reasoning strategy
• 💾 Persistence: SQLite storage for reasoning traces and analytics
• 📈 Usage Statistics: Track framework usage and effectiveness

Installation

git clone https://github.com/your-org/promptsmith
cd promptsmith
uv sync

MCP Tools

Usage

As MCP Server

Add to your MCP client configuration (e.g. .mcp.json in your project root):

{
  "mcpServers": {
    "promptsmith": {
      "command": "uv",
      "args": ["run", "--directory", "/path/to/promptsmith", "python", "-m", "promptsmith.main"],
      "env": {
        "PYTHONPATH": "/path/to/promptsmith/src",
        "UV_LINK_MODE": "copy",
        "FASTMCP_SHOW_STARTUP_BANNER": "false"
      }
    }
  }
}

Programmatic Usage

from promptsmith.domain import FrameworkSelector, PromptBuilder

# Analyze a task
selector = FrameworkSelector()
analysis = selector.analyze("Write a recursive function to calculate fibonacci numbers")

print(f"Category: {analysis.category}")               # code
print(f"Complexity: {analysis.complexity_score}")     # ~3.3
print(f"Framework: {analysis.recommended_framework}") # program_of_thoughts

# Generate a meta-prompt
builder = PromptBuilder()
result = builder.build(analysis.task, analysis=analysis)
print(result.meta_prompt)

Reasoning Frameworks

40 frameworks across 8 categories, selected automatically based on task type and complexity.

Architecture

src/promptsmith/
├── main.py              # MCP Server entry point
├── domain/
│   ├── frameworks.py    # 40 reasoning framework implementations
│   ├── selector.py      # Task analysis and framework selection
│   └── builder.py       # Meta-prompt assembly
├── persistence/
│   ├── models.py        # SQLAlchemy/Pydantic models
│   └── storage.py       # SQLite operations
└── utils/
    └── complexity.py    # Text complexity analysis

Integration with Other Agents

PromptSmith is designed to be called by other MCP-enabled agents.

result = mcp_call("promptsmith", "generate_meta_prompt", {
    "task": user_query,
    "context": relevant_context,
})
enhanced_response = llm.generate(result["meta_prompt"])

License

MIT