MCPHero Skills

Give your agent infinite tools — without infinite tokens.

Every time an agent calls an API, parses a database, or runs a calculation, it burns tokens on context, schemas, error handling, and output parsing. MCPHero flips this: the agent builds the tool once, deploys it as a persistent MCP server, and reuses it forever. A 50,000-token integration becomes a 50-token tool call on every subsequent run.

95-99% token savings. Self-evolving agents. Infinite tooling.

Works with Claude Code, OpenCode, OpenClaw, Cursor, and any agent that supports the Agent Skills spec.

How It Works

Without MCPHero                           With MCPHero
─────────────────                         ─────────────
Agent needs to query Postgres             Agent builds an MCP server
  → Reads schema docs (tokens)              with query tools (one-time)
  → Writes SQL (tokens)                     
  → Parses results (tokens)             Agent calls the tool forever
  → Handles errors (tokens)               → 1 tool call ≈ 50 tokens
  → Repeats EVERY run (tokens)            vs. 50,000+ tokens per run

Your agent doesn't just use tools — it creates them. Describe what you need in plain English, and MCPHero generates, deploys, and hosts a production MCP server with the exact tools your agent requires. The agent then registers the server and calls it on demand. No re-generation. No re-explanation. No wasted context.

This is how agents evolve: they start with general capabilities, build specialized tools as they work, and accumulate a growing toolkit that makes every future task cheaper and faster.

Available Skills

Prerequisites

Install mcpheroctl (for create-mcp-server skill)

# via Homebrew (macOS/Linux)
brew install arterialist/mcpheroctl/mcpheroctl

# via uv (cross-platform)
uv tool install mcpheroctl

Requires Python >= 3.12.

Authenticate

1. Log in to the MCPHero Dashboard.
2. Go to Settings → Organization → Developers.
3. Click Create API key.
4. Copy the generated token and run:

mcpheroctl auth login --token <YOUR_ORG_TOKEN>

Verify with:

mcpheroctl auth status

Config is stored at ~/.config/mcpheroctl/config.json.

Note: The meta-mcp-wizard skill does not require mcpheroctl — it works through the MCP protocol directly.

Installation

Option 1: CLI Install (Recommended)

Use npx skills to install skills directly:

# Install all skills
npx skills add arterialist/mcphero-skills

# Install a specific skill
npx skills add arterialist/mcphero-skills --skill create-mcp-server

# List available skills
npx skills add arterialist/mcphero-skills --list

# Install globally
npx skills add arterialist/mcphero-skills -g

Option 2: Claude Code Plugin

/plugin marketplace add arterialist/mcphero-skills
/plugin install mcphero-skills

Option 3: Manual Copy

git clone git@github.com:arterialist/mcphero-skills.git
cp -r mcphero-skills/skills/* ~/.claude/skills/

Usage

create-mcp-server (CLI workflow)

Ask your agent to build an MCP server using mcpheroctl:

"Create an MCP server for my PostgreSQL database with tools to query users and orders"
"Deploy a calculator MCP server with add, subtract, multiply, and divide tools"
"Help me set up an MCP server that wraps my REST API"

The agent walks through the wizard pipeline: describe requirements, suggest tools, configure env vars, generate code, deploy. Once deployed, the server persists — the agent (and you) can call those tools on any future run for a fraction of the token cost.

meta-mcp-wizard (Interactive MCP workflow)

Connect the Meta-MCP server to Claude Desktop or Cursor, then ask:

"Build me an MCP server for my Notion workspace"
"Create an MCP server that wraps our internal API"

The agent uses MCP tools directly — no CLI needed, no local setup beyond connecting the Meta-MCP server once.

License

MIT