cache.directory

What It Is

cache.directory is an AI tool registry “built by agents, for agents.” It catalogs agent skills, MCP servers, AI starters, system prompts, Cursor/Claude rules, local LLM tools, and more. It is not an open-submission directory: the site states that entries are researched and written by a content swarm, then validated against a strict Zod schema and reviewed through human curation.

Core Capabilities

Its core value lies in machine readability and reproducibility. Each entry uses Markdown plus YAML frontmatter, with fields such as name, description, author, last updated time, license, installation commands, and compatibility. Compatibility is marked for environments like Claude Code, Cursor, Cline, and Aider as full, partial, none, or untested. The site also provides /llms.txt and a static JSON API, making it easy for AI assistants, CLIs, and CI pipelines to consume directly.

API, CLI, and Integrations

cache.directory provides endpoints including /api/v1/skills.json, /api/v1/resolve/:slug.json, /api/v1/raw/:shelf/:slug, and /api/v1/search.json, with open CORS and no authentication required. @cache/cli can install entries via npx cache add and write them, depending on the shelf, to SKILL.md, .mcp.json, prompts, .cursorrules, or model files. It also generates a cache.lock with sha256 hashes, making it suitable for building reproducible Agent tool environments.

Security, Pricing, and Limitations

The security layer is relatively comprehensive: tools are checked in an isolated Docker environment for installation, behavior, unexpected outbound connections, file writes, and output consistency, and are assigned verdicts such as verified, caveat, suspicious, flagged, or untested. However, some examples in the crawled text are still marked untested, suggesting validation coverage may not yet be complete. On pricing, the page only states that there are no ads, sponsored listings, or paid placement; it does not disclose any plans or enterprise services.

Pros, Cons, and Who It’s For

Its strengths are a high degree of structure, API-friendly design, a simple CLI, and clear thinking around compatibility and reproducibility. Its drawbacks are that the current catalog is small, showing only 33 entries, and it lacks common enterprise SaaS features such as team permissions, SLA, audit logs, and SSO. It is best suited for AI Agent developers, engineering teams, and toolchain maintainers who need to discover, evaluate, and install prompts, skills, and MCP configurations.

Access from China

The crawled text does not provide information about mainland China network access, payments, or ICP filing status, so accessibility from China is unknown. If access to Cloudflare Pages or GitHub-related sources is unstable, alternatives include GitHub topic search, Awesome lists, Smithery, Glama MCP Directory, or an internal private tool directory maintained by your team.