How we evaluate

Every entry in AgentQuadrant carries a tier label that reflects how thoroughly we have checked it and how strongly we recommend it.

What the tiers mean

Editor's Pick is our highest designation. A tool earns it by being exceptional at what it does and by passing hands-on editorial testing. We actually install it, run it against real tasks, and verify that its documented behaviour matches reality. Editor's Picks are re-evaluated whenever a major version is released.

Verified means the tool has been checked: we have confirmed it installs cleanly, its source or listing is authentic, and its described functionality is substantiated by documentation or observable behaviour. Verified entries have not undergone the same depth of editorial testing as Editor's Picks, but they are not published on trust alone.

Listed means the entry has been indexed from a trusted source (the MCP Registry, the OpenAI GPT Store, or a public GitHub repository with a valid manifest) but has not yet received a deep editorial review. Listed tools are included so the directory is comprehensive; treat them as indexed, not yet endorsed.

Where our data comes from

MCP server data is sourced from the official Model Context Protocol registry and from public GitHub repositories that ship a marketplace.json manifest. Claude plugin and Skill data comes from public repositories containing valid SKILL.md or plugin frontmatter. Custom GPT data comes from the OpenAI GPT Store (public listings) and a licensed dataset of notable GPTs. Agent framework data is sourced from GitHub repositories with agent-specific frontmatter and from curated community lists with clear provenance.

We do not scrape marketing copy, ingest unverified community lists, or generate synthetic descriptions. All prose on an entry page is either quoted directly from the tool's own documentation (with attribution) or written by our editorial team after testing.

How often we re-verify

Each entry carries a last_verified date, visible in the entry footer. Our re-verification schedule is tier-dependent: Editor's Picks are checked on every major release; Verified entries are reviewed at least quarterly; Listed entries are swept periodically and either promoted or removed if they become unreachable or unmaintained. Entries that have not been re-verified within their expected window are flagged with a "stale" indicator until a review is completed. If you notice an entry that is out of date, use the submission form to flag it.