Live data · The Discry Index

The benchmark of how agents find and understand APIs.

Discry scores agent discovery and comprehension of APIs by giving AI models real tasks to complete against each API’s public docs.

395
APIs scanned
324
scored
75vs47
median comprehension vs discovery
Found and understood 113Understood, hard to find 14645%Found, misread 21Hard to find, misread 44
Discovery: can agents find it →Comprehension: can agents understand it →
APIs scored by agents
Explore the full Index →
Where APIs land

7% of scored APIs are Sharp. 38% are Murky or Opaque.

The Discry Score grades APIs into 5 bands, and breaks down discovery and comprehension scores. Learn more about the methodology.

  1. SharpAgents read these docs at full fidelity237%
  2. ClearReadable with minor friction7323%
  3. LegibleUsable, with real gaps10633%
  4. MurkyAgents misread these docs often8225%
  5. OpaqueEffectively unreadable to agents4012%
The gap, band by band

In most bands, the median API is easier to understand than to find.

Median discovery and median comprehension inside each band, with the distance between them. Read left to right: how far an agent gets before it can start reading. Scan your API to see which band it lands in.

23
point gap inside the Sharp band
  1. Sharp
  2. Clear
  3. Legible
  4. Murky
  5. Opaque
By category

AI leads. Infrastructure trails. Discovery lags comprehension in 11 of 11 categories.

Median scores per category, with the category leader. Open a leaderboard for the full ranking, or read the State of Agent-Readiness report for the story behind each category.

  1. AI44 scored · Leader: Firecrawl 96
  2. Productivity26 scored · Leader: Monday.com 88
  3. Payments30 scored · Leader: Razorpay 90
  4. CMS16 scored · Leader: Strapi 83
  5. Communication29 scored · Leader: Resend 93
  6. Commerce16 scored · Leader: akeneo.com 79
  7. Analytics25 scored · Leader: PostHog 85
  8. Auth & identity18 scored · Leader: Oso 89
  9. CRM18 scored · Leader: Attio 82
  10. Developer tools42 scored · Leader: CircleCI 88
  11. Infrastructure60 scored · Leader: Neon 88
Off the ladder

18% of the corpus was never scored. An agent could not reach the docs.

59 APIs are Invisible to agents: a plain HTTP fetch, the way most agents read, returned nothing usable. 11 more Walls agents out: the docs refuse automated access. Each one is listed with its reason, and each one is fixable. If yours is here, run a free scan once the docs are reachable.

Invisible to agents 59

Walls agents out 11

The discovery signals

The cheapest layer is the least built.

Share of the whole corpus publishing each machine-readable signal agents look for before they read a word of prose. Teams compete on comprehension quality. The files that get an agent to the docs are still rare. See who publishes them in the full Index.

  1. robots.txt87%343
  2. llms.txt75%297
  3. sitemap.xml72%285
  4. llms-full.txt47%187
  5. OpenAPI spec43%171
  6. Listed in an MCP registry30%120
  7. /.well-known/mcp.json12%48
  8. AGENTS.md3%12

of 395

The two MCP signals are separate probes: a listing in a public registry (Smithery) and a first-party manifest at /.well-known/mcp.json. 18 APIs have both.

Cost of comprehension

4% of scored APIs clear the small-model bar.

The capability floor is the cheapest tested model class that reliably completes tasks against the docs. Cost-based routing sends routine work to small models, so an API that needs a frontier model is expensive to use. 110 of 324 scored APIs have no floor reading yet and are shown as unmeasured, never as a floor. Scan your API to find its floor.

  1. Frontier required3410%
  2. Advanced-model required8627%
  3. Small-model ready144%
  4. Not agent-ready8025%
  5. Unmeasured11034%

Across 324 scored APIs with a reading, agents spent a median of 12,722 tokens per passed comprehension task, measured against claude-opus-4-8. Token cost moves with model version and prompt design. It is a diagnostic, never a score input.

What this page is

Discry measures documentation, not traffic. Every score comes from real models quizzed against an API’s public docs, fetched over plain HTTP with no JavaScript, and graded mechanically on instrument 3.0.1.

We do not host docs and we do not sit in the request path. Scoring is identical whether or not an API has ever paid Discry.

This view was generated Sep 7, 2026. It refreshes when the Index rescans.

Explore the IndexDiscry your APIRead the report →

Take the dataset: JSON · CSV