pipehero
Docs menu

Install the skill

A skill teaches your agent when and how to reach for Pipehero, on top of the MCP tools. Without it, the agent only has the tools' names and descriptions; with it, the agent also knows to infer the port and tunnel name from your repo, when to pass long_requests or realtime, and what a Pipehero-shaped request looks like ("expose my app for Stripe", "the last webhook 500'd — why?").

Install it

pipehero skill install            # ~/.claude/skills (all projects)
pipehero skill install --project  # ./.claude/skills (this repo only)

Prefer to do it by hand? It's the same content served at pipehero.app/skill.md — drop it into ~/.claude/skills/pipehero/SKILL.md.

What it covers

  • Exposing an app — inferring the port from .env, package.json or the framework's default, and a sane tunnel name from the repo.
  • Diagnosing a failed webhook — reading the captured payload and your handler together to explain why a request 500'd, fixing it, then replaying to confirm.
  • Exposing your own MCP server — knowing to pass long_requests: true so longer-running tool calls don't time out.
  • Testing on a real device via Realtime — and saying so up front if the workspace is on Free, instead of the tunnel silently failing.

FAQ

What's the difference between the skill and the MCP server?

The MCP server is the tools themselves — list_tunnels, get_request, replay_request and so on. The skill is instructions: it tells the agent when to reach for those tools (e.g. "expose my app for Stripe" → start_tunnel) and how to use them correctly (infer the port, pick a name, pass --long-requests for an MCP server, --realtime for mobile testing) without you spelling it out each time.

Do I need the skill if I already added the MCP server?

No, it's optional — the tools work either way. The skill just makes the agent proactive instead of reactive: it knows the example prompts and the right tool/flag for each, so you can ask in plain language instead of naming a tool.

What's the difference between installing globally and with --project?

Global (no flag) writes to ~/.claude/skills/pipehero/SKILL.md, so every project on your machine picks it up. --project writes to ./.claude/skills/pipehero/SKILL.md in the current directory instead, scoped to that one repo — useful if you only want it available (or checked into version control) for a specific project.

Does the skill file ever go stale?

It's embedded into the pipehero binary at build time from the same file served at pipehero.app/skill.md, so re-running pipehero skill install after an update overwrites it with the current version.