Another week, another sweep across the API Evangelist Network. This one was bigger than the last because I finished extending my blog pull to all 7,000-plus repos in my API Evangelist GitHub Organization — not just the 948 from the prior round. The coverage tripled: 6,518 posts across 2,364 API-bearing repos in the last seven days, 3,626 of those with API-related title signal, ranked and deduplicated into this week's top stories

Three threads defined the week. MCP Night 4 happened, and WorkOS turned the event into a one-day content blitz that defined the week's narrative. Multiple SaaS category I track shipped an MCP server — not just the developer-tools crowd; this wave is now hitting CRM, payments, scheduling, product management, and HR all at once. And for the first time, the conversation pivoted away from "here's our MCP server" toward "here's how to govern MCP in production." That last one is what I have been waiting for.

Let me walk through what I saw as Claude and I sifted through the noise.

MCP Night 4 Is An Agent-Ecosystem Conference

WorkOS published six recaps in a single day from MCP Night 4 on 2026-05-25, and each one is worth reading because they collectively sketch what the agent stack is going to look like by Q4.

The headline announcement is auth.md and "agentic registration" — a proposed .well-known/auth.md convention for how an agent discovers what a service expects for authentication. This is the obvious next primitive after .well-known/api-catalog and .well-known/mcp, and the fact that the conversation is now happening at the convention-protocol layer rather than the per-vendor SDK layer is a sign of maturity. I have been writing about .well-known patterns for years; agentic discovery has finally caught up.

AgentCard is the same idea applied to payments — one-time-use cards issued to agents for specific transactions, with policy enforcement at the card issuance layer rather than the merchant. This is going to be a major surface area. Agents need to spend money; the existing card network is not built for the "ephemeral, scoped, one-shot" issuance model that agents require. AgentCard is one of the cleaner answers I have seen.

AgentMail — email as an identity layer for agents — is the third piece in the same picture. If auth.md is "how does the agent know what auth method you use" and AgentCard is "how does the agent pay," AgentMail is "how does the agent receive and act on transactional state." Together these three primitives are the substrate that the next layer of agent infrastructure gets built on.

The panel recap on six months of agents is the most quotable. The summary if you want one line: the people building agents have moved from "can we make this work" to "how do we make this safe and predictable in production." The shift in tone from MCP Night 3 to MCP Night 4 was, by all accounts, significant.

The two demo recaps — Expo's Evan Bacon putting the iOS simulator in the browser and Rhys Sullivan's generative UI for agents — are the kind of "weird demo that turns into the obvious pattern in 18 months" that conferences exist for. Generative UI for agents in particular is going to be a thing. The agent doesn't render its own output; the output is rendered by a UI primitive the agent invokes as a tool. I expect to see this pattern shipped by multiple vendors by Q3.

If you are paying any attention to where the agent infrastructure conversation is going, MCP Night 4 was relevant. Six recaps in one day from one vendor (WorkOS) is also a tactic worth flagging — it positioned WorkOS as the organizing voice of the event, and the SEO surface they grabbed by being the only place writing up every demo is real.

Multiple SaaS Categories Shipped an MCP Server This Week

Last week I noted MCP gateways were entering API gateway territory. This week the pattern moved one layer up: every SaaS company is shipping an MCP server, and the wave has crossed out of developer tooling into operational SaaS.

Salesforce shipped the Data 360 MCP Server — and it landed in thirteen different repos in my network because Pardot, ExactTarget, MuleSoft, Heroku, and every other Salesforce property republished it. When the same announcement carpet-bombs your aggregator across 13 repos, the vendor is treating MCP as a flagship-level launch. Salesforce wants every analyst, every customer, every reseller to know they shipped this.

Zoho's MCP for payments is in the same neighborhood — agents acting against payment surfaces. Aha! (product management), Assembled (customer support ops), ChartMogul (revenue analytics, framed as "chat with your revenue data"), LeanData's BookIt MCP (revenue operations / scheduling), and Remote.com publishing nine internal use cases for their MCP all landed in the same seven-day window. Even Chainstack shipped a how-to for connecting their MCP server to Codex — the integration cookbook I called out from Merge.dev last week is now being copied by every infrastructure vendor.

The pattern is clear: if your SaaS product had an API in 2020, it now has an MCP server in 2026. If it didn't have an API in 2020, you are getting one bundled with the MCP server announcement this year, whether you wanted one or not.

The interesting question shifts to: which of these will actually be used at scale, and which are MCP-as-PR? My read is that the ones tightly coupled to a real underlying API (Salesforce Data 360, Remote, Zoho Payments) will see real adoption. The ones that are essentially "we wrapped our existing UI in MCP" will be quietly deprecated by Q1 2027. We are in the "everyone ships one to be in the press release" phase. Sustained usage comes later.

Governance Finally Showed Up

This is the one I have been waiting for. Portkey published What MCP Governance Actually Means in Production — the first piece I have read this year that treats MCP governance as a first-class operational concern rather than a "we'll figure it out later" footnote.

The framing is right: which MCP servers can your agents call, under what conditions, with what credentials, against what cost ceiling, with what audit trail, and who is accountable when the agent does something the policy didn't anticipate? The naive answer — "the MCP server is just an API, govern it the same way" — is not actually true once you have an agent in the middle, because the agent is making the tool-selection decision at runtime in a way that conventional API authorization doesn't model.

I expect a wave of MCP governance content from Kong, Apigee, Gravitee, MuleSoft, and the rest of the API management vendors over the next 30 days. The framing Portkey set up is going to be the framing everyone else adopts. The vendor that ships the cleanest "MCP governance from the gateway you already have" story is going to be quoted in every enterprise procurement conversation for the rest of 2026.

This is also where my Naftiko Signals work plugs in directly. Governance only works if you can see what is actually happening — which MCP servers exist, who is shipping new ones, what's coming online this week. The blog-pull pipeline I keep iterating on is one piece of that signal layer. The fact that I can now produce "every MCP-server announcement in the API Evangelist network in the last seven days" as a deterministic markdown index — that is exactly the kind of input MCP governance is going to need.

The "Turn Your Stack Into MCP" Tutorial Category Is Forming

APILayer published How to Turn Any REST API Into an MCP Server for Claude (Complete 2026 Pillar Guide) and they explicitly called it a "pillar guide" — meaning: this is the SEO anchor we are going to link every future MCP post back to. Smart positioning. I expect Postman, Speakeasy, Stainless, Stoplight, and every other API platform to ship their own version of this guide by mid-June.

Highlight.io shipped Auto-generating OpenAPI documents from TypeScript interfaces — same energy, different angle. "Your code already has the contract; we can extract OpenAPI from it; OpenAPI is the on-ramp to MCP." The flow from "TypeScript types → OpenAPI → MCP server" is now a recognized pipeline.

Bindbee shipped five integration-explainer posts in a couple of days — B2B API integration best practices, API integration benefits, unified API strategies, API-led integration concepts, and a "best API integration tools in 2026" comparison roundup. This is the same content-marketing playbook Merge.dev ran last week with their MCP-cookbook series. The "I am the integration platform; here is the SEO surface for buyers Googling 'API integration' in 2026" play is going to be every iPaaS vendor's Q3 strategy.

A Few Other Threads Worth Noting

Anyscale published "Reimagining ML Operations with Agent Skills: a new maturity model for on-call" — Agent Skills are now being pitched as the new MLOps primitive. This is a notable expansion of the Agent Skills surface area; Anthropic introduced Skills as a developer-side capability, and Anyscale is now pitching them as the operations-side primitive too. Watch this thread.

Wowza's "Video API vs Video SDK" is a small story but a useful one — the API-vs-SDK comparison piece is a sub-genre that gets written every time a vendor wants to clarify when each form factor wins. This will get more frequent as agents force a re-evaluation of what an SDK is even for.

Photoroom shared a customer story — Completeful saw a 100% sales boost after integrating Photoroom's API. The API-as-business-multiplier narrative is alive. Worth flagging because it is one of the cleaner "API generated X dollars" case studies I have seen this year, and it landed in my network through a corporate blog feed, not a PR push.

Cognism shipped "What is Technographic Data? API & Targeting Guide" — a small thing, but the technographic-data category has been quietly heating up as B2B sales tooling pivots to agent-driven outbound. Worth watching.

On API Evangelist

The Fortune 1000 AI investment report I shipped earlier this week — the one where I scored all 989 Fortune 1000 companies that I have repos for against a 35-term AI lexicon — is now live and rankable. American Express, Hanover Insurance, and Teradata lead the normalized signal density rankings; Alphabet, Salesforce, and Coca-Cola lead the press-release-AI-coverage rankings. The full per-company markdown reports sit next to each repo's apis.yml. The companion blog post frames the headline finding: zero of 989 companies score zero on AI signal, which means "is your industry talking about AI yet" is now the wrong question.

The blog-pull pipeline I have been iterating on hit a meaningful milestone this week. It now covers 3,166 repos across the API Evangelist network with at least one blog post pulled and indexed — up from 979 last week. This is becoming the primary input to the weekly newsletter, and the more repos I have feeding it, the better the signal gets.

What I Am Watching Going Into Next Week

Three things.

The MCP governance content wave. Portkey set the framing on 2026-05-25. I expect Kong, Apigee, MuleSoft, Gravitee, and Solo.io to ship their own "MCP governance with the gateway you already have" pieces over the next two weeks. The vendor that ships the cleanest authoritative version wins the enterprise conversation.

The agentic primitives (auth.md, AgentCard, AgentMail) becoming real conventions. WorkOS proposed them at MCP Night 4. The question now is whether the rest of the ecosystem (Anthropic, OpenAI, the integration platforms, the identity providers) adopts the conventions or proposes their own. If consensus forms around .well-known/auth.md, that is going to be one of the most important developments of 2026.

The Salesforce Data 360 MCP rollout. Salesforce got 13 repos in my network to syndicate the announcement on day one. Watch how their network plays this through Q3. If Data 360 MCP actually works at scale, the enterprise CRM/data stack conversation will work to anchor on it.

Coming soon: I will also start publishing the same kind of weekly newsletter at the area, stage, and experience level — the stories that don't make this newsletter still get organized and routed to the APIs.io blog and to my LinkedIn, Bluesky, and other channels. The point of the network is that nothing interesting falls through the cracks; the newsletter is just the part that surfaces. I am even considering evolving the blog here to handle the scope.

If you are wondering how I am producing this every week without losing my mind: I am not doing it by hand. The pipeline is deterministic and runs in three skills — pull-blogs to pull RSS across the network, blog-pull-reports to produce the five standard reports, and then a writeup pass where Claude and I work through the top stories together. I will keep showing the work as I iterate on it. The point of doing this in the open is that the methodology is the product — anyone can run the same pipeline against their own list of providers, their own portfolio, their own competitors, and produce the same kind of weekly read. I am not sure where where all of this is going and I am just working to iterate upon it each week.