Cloudflare and Microsoft Azure API Management rank first and second on the AI gateway tag, and both are exemplars: Cloudflare at 71.1, Azure API Management at 67.3. The Kin Score measures the public API surface a company publishes, not the product or the company behind it. The tag is small and strong, 33 providers averaging 51.4 against 22.6 for the rest of the catalog. It is also two things at once. Most of its members route and observe LLM traffic, and about a quarter are enterprise API gateways that learned to do the same. These two arrived at the AI gateway from opposite ends, and the scores show which end each one came from.
An edge network against an API management platform. Cloudflare sells a global network, and its AI Gateway is one product among dozens on it, next to Workers, R2, D1, the WAF, DNS and Zero Trust. Its access model is freemium and self-serve, with six documented plans, from a free plan with unlimited bandwidth to Pro at $25 per zone per month and Workers Paid at $5 per month. It also documents a global REST cap of 1,200 requests per five minutes. Azure API Management is a hybrid, multicloud gateway with a management plane and a developer portal. Its AI gateway is a set of capabilities layered on that gateway: token rate limiting, semantic caching, content safety, and load balancing across Azure OpenAI, OpenAI, Anthropic, Foundry and Bedrock backends, plus MCP servers and A2A agent APIs. It is also freemium and self-serve, with eight plans on record. If you want an AI gateway in front of your model calls, Cloudflare is the lighter purchase. If you already govern hundreds of APIs and want LLM traffic under the same policies, Azure API Management is the one that fits.
Cloudflare ships breadth, and Azure ships depth on one resource model. Cloudflare’s 53 OpenAPI contracts carry 1,621 operations across 84 API pages, from the WAF and Workers to AI Search and Agents. The gateway itself surfaces as a five-operation Gateways API and an OpenAI-compatible API. Cloudflare also runs 14 MCP servers, publishes 11 agent skills, and ships two AsyncAPI documents and a GraphQL analytics API. Azure API Management’s 74 contracts carry 307 operations across 71 pages, nearly all split from the 2024-05-01 Resource Manager specification Microsoft publishes, covering the service, API policies and API Center. One disclosure belongs here. Ten of Azure’s 74 contracts were derived by us from documentation, and the AI gateway contract is one of them. Microsoft configures its AI gateway through policies on the management plane and publishes no separate contract for it. Nothing in this post credits Microsoft for that page.
Three facets each, and Cloudflare holds the heavy ones. Cloudflare takes contract quality 76.5 to 58.5, developer ergonomics 96.4 to 89.3, and access clarity 84.2 to 76.3. Those three carry 0.65 of the weight. Contract quality is the biggest gap at 0.25. Cloudflare publishes OpenAPI 3.1 and AsyncAPI, while Azure’s contracts are OpenAPI 3.0 and declare the implicit OAuth grant that current guidance deprecates. Azure takes discoverability 77.8 to 57.4, because Cloudflare’s catalog record carries a one-line description and no base URL on every API. It also takes contract governance 28.8 to 13.6 and operational transparency 52.6 to 44.7. Neither of those two leads is Microsoft’s own work. The whole governance gap is a single check for a published vocabulary, and the vocabulary file on Azure’s record was generated by our pipeline. The transparency gap is mostly one four-point check for a security disclosure route. That check reads a link typed Security, and Azure’s link of that type is an authentication overview page. Cloudflare records a real vulnerability disclosure under a different type, and the check does not count it. Both defects are ours, and both are filed. The six facets weight out to 68.4 against 65.8. The conditional layers, led by create-or-update ergonomics, add 2.7 to Cloudflare and 1.5 to Azure, and that makes the 3.8 gap.
On agents, the distance is wider than the composite. Cloudflare scores 65.8 and is banded agent-ready. Azure API Management scores 27.5 and is agent-aware, 1.1 points under the agent-ready floor of 28.6. Cloudflare’s score clears the 38.7 floor for agent-native, but the band is gated. Agent-native also requires documented idempotency and a stable error envelope. Cloudflare verifies its errors and documents no idempotency, so it stops one band short. Cloudflare alone lights error semantics, agent skills, an agent card, delegated identity, protected resource metadata and dynamic client registration. Both publish MCP and document reversibility and rate limits. Neither offers idempotency, a dry-run mode, a well-known catalog or consent identity. The tag leads the rest of the catalog on MCP, 37.5% to 12.6%, and still documents idempotency for only 9.4% of its members.
Pick on where your traffic already goes. If you are putting a gateway in front of model calls and want an agent to operate it, Cloudflare’s surface is the stronger one on every facet that carries real weight. It is also the only one of the two an agent can authenticate to without a human registering a client first. If your APIs already run through Azure and you want one policy layer over REST, LLM, MCP and A2A traffic, Azure API Management’s surface is the one you will live in, and its management contracts come straight from Microsoft. Each side has a clear path to a higher number. Azure moves most by publishing its AI gateway as a contract of its own, moving its specifications to OpenAPI 3.1 without the implicit grant, and documenting error semantics, which on the published weights is worth more than the 1.1 points it needs to reach agent-ready. Cloudflare moves most by documenting idempotency, which would make it agent-native, and by giving its catalog record a real description and a base URL on every API.