Agent-readiness score
Every company in the directory carries a 0–100 agent-readiness score: a measure of whether AI agents can discover, understand, and operate the product. This page is the reference for how it's computed.
The six signals
| Signal | Points | Probe |
|---|---|---|
| Public API | 30 | Documented HTTP API reachable by an agent — detected via docs pages, OpenAPI specs, and homepage claims we verify |
| MCP server | 25 | A Model Context Protocol endpoint, advertised in docs or llms.txt |
| Webhooks | 15 | Outbound event delivery (agents subscribe instead of poll) |
| OAuth | 10 | Delegated authorization flow an agent can complete on a user's behalf |
| API docs | 10 | Machine-readable reference — /openapi.json or /.well-known/openapi.json score highest |
| SDKs | 10 | Official client libraries (any mainstream language) |
Points sum to a 0–100 score. Sub-scores are shown on each startup profile under the agent-readiness panel.
Grading bands
- 70–100 · Agent-ready — an autonomous agent can meaningfully use the product today. These companies appear on the agent-ready leaderboard.
- 40–69 · Partially ready — discoverable, but missing pieces (usually webhooks or machine-readable docs) force human intervention.
- 0–39 · Not ready — agents can read the marketing site at best.
Related probes: the site scan
The public checker also probes discoverability signals that inform (but don't directly sum into) the profile score: /llms.txt presence, explicit AI-crawler rules in robots.txt (GPTBot, ClaudeBot, PerplexityBot, Google-Extended, CCBot), /.well-known/ai-plugin.json, and API/MCP mentions on the homepage. Files that return HTML masquerading as the requested resource are rejected.
Corrections and rescans
Scores refresh when a profile is re-enriched. If your score is stale — you shipped an MCP server yesterday — claim your profile and request a rescan from the profile page, or file a correction. Manual overrides exist but are used only to fix probe errors, never as a favor.
Why it exists
Agents are becoming a distribution channel. When a user asks an assistant "find me a transcription API and set it up," the assistant can only pick products it can actually operate. The score makes that gap visible — and the agent-ready guide shows how to close it.