metaskill MCP — Use Cases, Install & Live Demo

Why use it

Key features

/metaskill single command — no manual role design
Research phase runs live web searches so output reflects current ecosystem
Generates routing doc, agents, skills, standards, and MCP config in one pass
Three intent modes: full team / single agent / single skill
Default architecture uses Opus tech-lead + Sonnet specialists with review gates

Live Demo

What it looks like in practice

metaskill-skill.replay ▶ ready

0/0

Install

Pick your client

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "metaskill-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/xvirobotics/metaskill",
        "~/.claude/skills/metaskill"
      ],
      "_inferred": true
    }
  }
}

Open Claude Desktop → Settings → Developer → Edit Config. Restart after saving.

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "metaskill-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/xvirobotics/metaskill",
        "~/.claude/skills/metaskill"
      ],
      "_inferred": true
    }
  }
}

Cursor uses the same mcpServers schema as Claude Desktop. Project config wins over global.

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "metaskill-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/xvirobotics/metaskill",
        "~/.claude/skills/metaskill"
      ],
      "_inferred": true
    }
  }
}

Click the MCP Servers icon in the Cline sidebar, then "Edit Configuration".

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "metaskill-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/xvirobotics/metaskill",
        "~/.claude/skills/metaskill"
      ],
      "_inferred": true
    }
  }
}

Same shape as Claude Desktop. Restart Windsurf to pick up changes.

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "metaskill-skill",
      "command": "git",
      "args": [
        "clone",
        "https://github.com/xvirobotics/metaskill",
        "~/.claude/skills/metaskill"
      ]
    }
  ]
}

Continue uses an array of server objects rather than a map.

~/.config/zed/settings.json

{
  "context_servers": {
    "metaskill-skill": {
      "command": {
        "path": "git",
        "args": [
          "clone",
          "https://github.com/xvirobotics/metaskill",
          "~/.claude/skills/metaskill"
        ]
      }
    }
  }
}

Add to context_servers. Zed hot-reloads on save.

claude mcp add metaskill-skill -- git clone https://github.com/xvirobotics/metaskill ~/.claude/skills/metaskill

One-liner. Verify with claude mcp list. Remove with claude mcp remove.

Use Cases

Real-world ways to use metaskill

Bootstrap an AI agent team for a fullstack project

👤 Founders and solo devs who want structured multi-agent setups ⏱ ~20 min beginner

When to use: Starting a new project and you'd like proper agent separation instead of one overloaded Claude.

Prerequisites

Skill cloned — git clone https://github.com/xvirobotics/metaskill ~/.claude/skills/metaskill

Flow

Describe the project

/metaskill fullstack web app with React, TanStack, Postgres, Stripe subscriptions.✓ Copied

→ Research phase runs web searches; then build phase produces routing + agents + skills
Review the generated team

Walk me through each agent's scope and the routing rules between them.✓ Copied

→ Clear per-agent responsibility with no overlap
Provide credentials (optional)

Skip credentials for now — I'll add them later.✓ Copied

→ Skill proceeds to verification without blocking

Outcome: A working multi-agent setup you can start using immediately.

Pitfalls

Generated team has too many roles for a 1-person project — Use the single-agent mode instead ('create a single agent for X')

Combine with: github · filesystem

Create one specialized agent for an existing workflow

👤 Devs who already have a project and want to add one role ⏱ ~10 min beginner

When to use: You need a 'security reviewer' agent but already have everything else.

Flow

Invoke single-agent mode

/metaskill — a single security reviewer agent for a Node.js API.✓ Copied

→ Targeted agent file with scope, not a full team

Outcome: One focused agent integrated into your existing setup.

Generate a single deploy skill without building a whole team

👤 Devs formalizing a workflow ⏱ ~15 min beginner

When to use: You keep running the same deploy steps manually and want a skill for it.

Flow

Invoke single-skill mode

/metaskill — a deploy-to-staging skill for Cloudflare Workers with rollback on health-check failure.✓ Copied

→ SKILL.md with triggers and steps

Outcome: A reusable skill you can install right away.

Combine with: skillcheck-free-skill

Combinations

Pair with other MCPs for X10 leverage

metaskill-skill + skillcheck-free-skill

Validate Metaskill's generated SKILL.md files against the agentskills spec before using them

After /metaskill, run SkillCheck on every generated SKILL.md.✓ Copied

metaskill-skill + github

Commit the generated team structure as a first PR to seed your repo

After generation, commit .claude/ and agent docs in a single 'seed' PR.✓ Copied

Tools

What this MCP exposes

Tool	Inputs	When to call	Cost
generate_team	project description	Fresh project	0 + web search tokens
generate_single_agent	role description	Existing team, need one more role	0
generate_single_skill	workflow description	Formalizing a repeated workflow	0
verify	generated files	Auto-run after generation	0

Cost & Limits

What this costs to run

API quota: Uses web search during research phase
Tokens per call: High for team generation (research + build); lower for single-agent/skill modes
Monetary: Free — skills are local files. Web search may count against your own MCP's quotas.
Tip: Use single-agent mode when you don't need a full team; it's much cheaper.

Security

Permissions, secrets, blast radius

Credential storage: No credentials in the skill itself. Credentials phase asks for your API keys if you opt-in.

Data egress: Research phase performs web searches

Generated agents are suggestions — review them before granting them MCP tool access.

Troubleshooting

Common errors and fixes

Skill not invoked — Claude sets up agents manually

Use '/metaskill' as a literal trigger.

Verify: ls ~/.claude/skills/metaskill/SKILL.md

SKILL.md frontmatter wrong

Re-clone; description drives activation.

Verify: head ~/.claude/skills/metaskill/SKILL.md

Wrong trigger keywords — generates a skill when you wanted a team

Phrase the project description clearly; 'fullstack web app' triggers team mode, 'a skill for X' triggers skill mode.

Alternatives

metaskill vs others

Alternative	When to use it instead	Tradeoff
Manual agent design	You have strong opinions about role structure	More control, much slower
rootnode-skill	You want Project-level engineering rather than agent-team generation	Different axis — Claude Projects vs Claude Code multi-agent

More

Resources

📖 Read the official README on GitHub

🐙 Browse open issues

🔍 Browse all 400+ MCP servers and Skills