Tools Info Platform MCP Server

An extremely fast, lightweight Model Context Protocol (MCP) server that holds and manages rich, structured "paradigmatic" documentation pages for software tools, packages, and frameworks. It uses DuckDB as a superfast, serverless analytical storage engine.

This platform is coupled with the paradigmatic-docs Agent Skill, designed to convert any software tool's raw documentation or instructions into a high-quality developer-centric instruction blueprint covering: 1. Application Goals (Core Purpose, Target Audience, Key Use Cases) 2. Qualities & Design Decisions (Architecture, Dependencies, Extensibility) 3. Audits & Reliability (Security Profile, Error Handling, Health Checks) 4. Metrics & Performance (Throughput, Memory Footprint, Scaling) 5. Advantages & Limits (Pros & Cons) 6. Installation & Setup 7. Paradigmatic Code Examples (Complete, robust, runnable scripts and production setups)

The server ships pre-seeded with the ultimate gold standard template: MLX-VLM (Vision-Language Models optimized for Apple Silicon with continuous batching, speculative decoding, and automatic prefix caching).

Architecture & Storage Design

The platform uses DuckDB as its storage engine. This choice brings several advantages: - Serverless & Local: Data resides in a single, high-performance file (tools_info.duckdb), requiring zero database administration, network configurations, or background processes. - Superfast Access & Full-Text Search: Instantly filters, retrieves, and processes large pieces of markdown, JSON structure, or logs. - Relational & Analytical Power: Exposes a safe, read-only SQL execution tool, allowing agent-based systems to perform counts, listings, category groups, or complex regex filtering inside the database.

Schema

CREATE TABLE tools_docs (
    name VARCHAR PRIMARY KEY,
    title VARCHAR,
    category VARCHAR,
    description TEXT,
    goals TEXT,
    qualities TEXT,
    audits TEXT,
    metrics TEXT,
    advantages TEXT,
    limits TEXT,
    installation TEXT,
    usage_examples TEXT,
    raw_markdown TEXT,
    updated_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
);

Exposed MCP Tools

The server runs on stdio transport, exchanging JSON-RPC 2.0 messages. It exposes the following tools:

Installation & Setup

1. Install Dependencies: Make sure you have python and duckdb installed in your system or virtual environment: bash pip install duckdb (Note: This MCP server uses built-in Python standard libraries for standard input/output parsing, meaning no external MCP package is strictly required, making it highly robust).

2. Starting the Server: bash python3 server.py

Adding the MCP Server to Zed / Claude Desktop

You can easily register this server in your Zed editor or Claude Desktop configuration file (~/.config/zed/settings.json or ~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcp": {
    "servers": {
      "tools-info-platform": {
        "command": "python3",
        "args": [
          "/Users/friedrichstr/ai-mac-studio/apps/tools-info-platform/server.py"
        ]
      }
    }
  }
}

Pre-seeded Golden Standard: MLX-VLM

When first started, the database auto-populates with mlx-vlm. This document contains comprehensive, robust instructions on running local multimodal LLMs on Apple Silicon: - Standard image description scripts - Multi-modal scripts (combining audio and image models) - Server commands and options (thinking budgets, speculative decoding drafters, and trust-remote-code settings) - Automated Prefix Caching (APC) Python scripts and Server flags - KV cache quantization (TurboQuant 3.5-bit or uniform 8-bit)