# Scrapyard

> Scrapyard is an Apache-2.0 self-hosted REST and MCP service for web search and clean page extraction. It runs on the user's machine, requires no account or API key, and sends no telemetry.

Scrapyard supports Python 3.12. Install it with `python3.12 -m pip install scrapyard`, install the required Chromium runtime with `python3.12 -m playwright install chromium`, then run `scrapyard`. The service binds to `127.0.0.1:8377` by default and creates a local authentication token on first run.

The REST API exposes `/search`, `/extract`, `/map`, `/research`, and `/status`. The stdio MCP server exposes the same core functions to Claude Code, Claude Desktop, Cursor, Hermes, and other MCP clients. Extraction responses report provenance including method, pipeline steps, upstream status, and final URL.

Limits: requests originate from the user's IP; there is no proxy pool or site-wide crawl; concurrency is 3; Google search is deliberately absent because it challenges non-browser clients. Scrapyard does not claim guaranteed access to defended sites.

The maintenance workflow is assist-only: a daily canary detects broken search adapters and produces a repair brief for the user's coding agent. A repair is not accepted until the security suite, canary, and live verification gate pass.

## Primary resources

- [Source and README](https://github.com/artboarding-hash/scrapyard): Canonical source, installation, operation, security model, and limits.
- [MCP setup](https://github.com/artboarding-hash/scrapyard/blob/main/docs/MCP.md): Configuration for MCP-capable clients.
- [Security policy](https://github.com/artboarding-hash/scrapyard/blob/main/SECURITY.md): Threat model and vulnerability reporting.
- [License](https://github.com/artboarding-hash/scrapyard/blob/main/LICENSE): Apache License 2.0.

## Package and discovery

- [PyPI](https://pypi.org/project/scrapyard/): Published Python package and release versions.
- [Official MCP Registry](https://registry.modelcontextprotocol.io/): Search for the published Scrapyard server entry; the registry entry is the canonical MCP discovery record once published.