# Pipehero — MCP server + skill

[**Pipehero**](https://pipehero.app) tunnels your `localhost` to a public URL and captures
every webhook, so you can **inspect, replay, and debug** them — including from your AI
agent via MCP.

> This repo is auto-published from Pipehero's source. It holds the public MCP manifest
> and the agent skill. Issues/PRs welcome.

- 🌐 Website: https://pipehero.app
- 📚 Docs: https://pipehero.app/docs

## Connect the MCP server

Let an agent (Claude Code, Cursor, Copilot, Windsurf, Zed, …) list your tunnels, read
captured webhooks, and replay them to localhost.

### Remote — recommended (no install, OAuth)

```bash
claude mcp add --transport http pipehero https://mcp.pipehero.app/mcp
```

Your client opens the browser to sign in and approve — no token to paste.

For CI/headless, pass a token from https://pipehero.app/dashboard/settings:

```bash
claude mcp add --transport http pipehero https://mcp.pipehero.app/mcp \
  --header "Authorization: Bearer <token>"
```

### Local — via the CLI

```bash
curl -fsSL https://pipehero.app/install | sh
pipehero login
claude mcp add pipehero -- pipehero mcp
```

Any MCP client works — add it to `.cursor/mcp.json`, `claude_desktop_config.json`, etc.:

```json
{ "mcpServers": { "pipehero": { "command": "pipehero", "args": ["mcp"] } } }
```

## Tools

| Tool | What it does |
|---|---|
| `list_tunnels` | Your tunnels and whether each is online. |
| `list_requests(subdomain)` | Recent captured webhooks for a tunnel. |
| `get_request(subdomain, id)` | Full request + response (headers + body). |
| `replay_request(subdomain, id)` | Replay a webhook to your localhost. |
| `start_tunnel(name, port)` | Expose a local port on a public URL (local MCP only). |
| `stop_tunnel(name)` | Stop a tunnel started with `start_tunnel` (local MCP only). |
| `list_api_collections`, `get_api_collection(collection_id)` | Your saved API docs: folders, endpoints, environments (remote MCP, and local from 0.1.14). |
| `get_api_endpoint(collection_id, method + path)` | One endpoint in full, with its version and response examples. |
| `upsert_api_endpoint(collection_id, method, path, …)` | Create or update an endpoint by route, so docs follow your code. |
| `add_api_example(collection_id, method + path, name, status, …)` | Save a response example. |
| `delete_api_endpoint(collection_id, method + path)` | Remove an endpoint whose route is gone. |
| `import_api_spec(format, content)` / `export_api_collection(collection_id)` | OpenAPI, Postman and curl, both ways. |
| `run_api_endpoint(collection_id, method + path, environment)` | Send it and read the real response. |
| `save_captured_request(collection_id, tunnel, request_id)` | Turn a real captured request into docs, without credentials. |
| `check_api_drift(collection_id, tunnel)` | Compare the docs with real traffic (Pro/Team). |
| `list_api_monitors(collection_id)` | Which endpoints are monitored and which are failing, with the last error (Pro/Team). |
| `list_api_proposals` | What happened to your proposed changes, on workspaces that review AI changes. |
| `get_api_share_link(collection_id, method + path)` | The endpoint's shareable link: a read-only page with its docs, examples and latest run. |
| `share_api_endpoint(collection_id, method + path, …)` | Create that link (workspace-only by default), open it to anyone with the link, or invite people by email. |
| `share_api_collection(collection_id, …)` | The same for a whole collection: every endpoint, read-only, never its environments. |

Because your agent has both the captured webhook and your codebase, it can explain *why*
a handler failed — then fix it and replay to confirm.

## Skill

Install the skill so your agent knows *when and how* to use Pipehero:

```bash
pipehero skill install
```

or copy [`SKILL.md`](./SKILL.md) into `~/.claude/skills/pipehero/SKILL.md`.

## License

MIT