Claude Meta Dorker

Claude-Meta-Dorker is a repository of per-service search manifests that describe the advanced-search URL surface of public websites in a form intended for AI agents to parse. Each manifest records the query parameters, inline operators, URL construction grammar, known constraints such as rate limits, authentication requirements and result caps, and provenance indicating when and how the surface was last verified. The project deliberately treats human readability as a non-goal, prioritising declarative, agent-oriented structure over prose documentation. It explicitly excludes general Google dork lists and OSINT cheatsheets, and it does not perform crawling or scraping, leaving execution to the consuming agent.

The manifests are written as YAML and validated against a JSON Schema, with Playwright-based validators intended to live-check that constructed URLs behave as documented. Before a manifest is written, a surface reconnaissance pass enumerates every search-like interface a service exposes, including official REST or GraphQL APIs, RSS and Atom feeds with query parameters, sitemap-driven search, opensearch.xml descriptors, and third-party mirrors offering stronger search. Where a non-UI surface covers the need more reliably, the manifest directs the agent to that surface instead of duplicating UI documentation. The repository is organised into directories for manifests, schema, agent guides, validation, and reconnaissance notes, and is released under the MIT licence.

The project is aimed at developers building AI agents or tool-using language model workflows that need reliable, non-trial-and-error methods for querying specific websites. It may also interest researchers studying agent tooling and maintainers who wish to expose structured search guidance for their own services. The project is currently in a baseline stage, and a Claude Code plugin is planned once the manifest schema and a stable set of services are established.