Skills & MCP

Skills give your agent reusable expertise. MCP servers give it new tools. minnal loads both lazily so they cost nothing until you need them.

Skills

A skill is a versioned bundle of instructions and (optionally) helper scripts that teach the agent how to do something well — a coding standard, a deploy workflow, a domain DSL. Skills are not all loaded on startup; the conversation is embedded as a vector and a skill is auto-injected if the embedding matches.

Install a skill

# From a git URL
minnal skill add https://github.com/you/my-skill

# From a local path
minnal skill add ./skills/my-skill

Skill layout

my-skill/
├── skill.json         # name, description, triggers
├── README.md          # human docs
└── prompts/
    └── system.md      # injected into the system prompt when active

Activate a skill

/skills              # picker
/skill my-skill      # activate by name

MCP (Model Context Protocol)

minnal speaks MCP natively. Add servers to ~/.minnal/mcp.json (or per-repo at .minnal/mcp.json):

{
  "servers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/me/projects"]
    },
    "github": {
      "command": "gh-mcp-server",
      "env": { "GITHUB_TOKEN": "${env:GH_TOKEN}" }
    },
    "postgres": {
      "command": "mcp-server-postgres",
      "args": ["postgres://localhost/dev"]
    }
  }
}

Manage MCP servers

minnal mcp list
minnal mcp add <name>
minnal mcp remove <name>
minnal mcp logs <name>

Agent grep

minnal ships a smarter grepbuilt for agents. It augments results with file structure (function list, displacement) so the agent can infer a file's shape without reading it, and it adaptively truncates repeated hits to save context.

Custom tools

• Drop an executable into ~/.minnal/tools/ with a tool.json manifest to expose it as a callable tool.
• Use a Skill to bundle a system prompt, a tool, and a recommended model together.