aptitude-resolver 0.1.4

Deterministic package-manager-style resolver for AI skills

Install

Install the published package as a CLI tool:

uv tool install aptitude-resolver
aptitude --help

Run it without a persistent install:

uvx aptitude-resolver --help

The package installs these console commands:

• aptitude
• aptitude-resolver
• aptitude-mcp

aptitude is the command most users should run. For MCP hosts that should launch the published package without a persistent install, use uvx aptitude-resolver mcp.

Configure Registry Access

Registry-backed commands need an Aptitude Server URL and read token:

export APTITUDE_SERVER_BASE_URL=https://your-aptitude-server.example
export APTITUDE_READ_TOKEN=your-read-token

For local development against the default server:

export APTITUDE_SERVER_BASE_URL=http://localhost:8000
export APTITUDE_READ_TOKEN=reader-token

aptitude policy show can still inspect local policy without contacting the registry.

Usage

Launch the install-first wizard:

aptitude

Install a skill by query:

aptitude install "Postman Primary Skill"

Inspect effective local policy and config layers:

aptitude policy show

Replay an existing lockfile:

aptitude sync --lock aptitude.lock.json

Show the full command and flag surface:

aptitude manifest

MCP Server

Aptitude ships a local stdio MCP server for agents and MCP-compatible apps. The recommended MCP host configuration runs the published PyPI package locally through uvx:

{
  "mcpServers": {
    "aptitude": {
      "command": "uvx",
      "args": [
        "aptitude-resolver",
        "mcp"
      ],
      "env": {
        "APTITUDE_SERVER_BASE_URL": "http://localhost:8000",
        "APTITUDE_READ_TOKEN": "your-local-read-token"
      }
    }
  }
}

The MCP server process still runs locally. uvx only resolves the published package and starts its mcp command over stdio.

The MCP server exposes tools for search, inspect, resolve, policy inspection, install, and lock sync. Mutating tools require explicit filesystem targets.

If Aptitude is already installed as a persistent tool, MCP hosts can also use the direct executable:

{
  "mcpServers": {
    "aptitude": {
      "command": "aptitude-mcp",
      "args": [],
      "env": {
        "APTITUDE_SERVER_BASE_URL": "http://localhost:8000",
        "APTITUDE_READ_TOKEN": "your-local-read-token"
      }
    }
  }
}

Replace APTITUDE_READ_TOKEN with a token accepted by the target Aptitude Server.

What Works Today

• discovery-backed query resolution
• candidate version selection
• recursive dependency graph resolution
• policy filtering and graph governance
• lockfile generation and replay
• archive-based skill materialization from verified tar.zst artifacts
• local config loading from aptitude.toml
• CLI and MCP interfaces

Source

Source and contributor documentation live in the project repository:

https://github.com/aptitude-stack/resolver