Methodology
Each product record answers one question — can AI agents automate this software? — per modality: API, SDK, MCP servers, no-code integrations, CLI, extensibility, data access, RPA/UI automation, and computer use.
Evidence tiers
| Tier | Meaning |
|---|---|
| declared | The vendor's own documentation says so; linked. |
| scraped | A registry or listing (MCP registries, marketplaces) says so; linked. |
| community | Practitioner reports — forums, blog posts, issue trackers; linked. |
| measured | We ran the probe ourselves; the record links the artifact. Computer-use verdicts are only ever this tier. |
Honest unknowns
Absence of evidence is never treated as evidence of absence. A modality nobody has documented renders as unknown — which tells you more than a guessed score would.
Scores
0–10 per modality plus an overall judgment of the best realistic agent path today, under a versioned rubric; records carry the rubric version they were scored under, so re-scores stay comparable.