claude marketplace: which of the three layers you can actually use.

elisabeth hitz · published september 24, 2026 · updated september 24, 2026 · 6 min read

anthropic opened a marketplace yesterday. most of the coverage read it as a store you can shop in. two of its three layers are not open to you, and the one that is has a running cost nobody put in the announcement.

anthropic launched claude marketplace on september 23, 2026 with three layers: connectors and plugins, agents and products, and service partners. only the first is reachable without an enterprise spend commitment. the other two are bought with a portion of your committed anthropic spend. if you run a one-person business, one third of that announcement is for you.

that is not a complaint. it is a filter. the fastest way to waste a week is to read an enterprise procurement announcement as a to-do list. so here is the split, what each layer costs, and the five checks worth running before you install anything from the layer you can actually reach.

the three layers, and who each one is for

the marketplace bundles three different things that happen to share a page: extensions you install into claude, third-party software built on claude that you buy, and humans you hire to roll claude out. anthropic put more than 2,000 connectors and plugins in the first layer, including atlassian, google, microsoft, notion, and salesforce (anthropic, sep 23 2026).

the layerwhat is in itwhat it costs youreachable solo?
connectors and pluginsmore than 2,000 today, including atlassian, google, microsoft, notion, salesforce (anthropic, sep 23 2026)no spend agreement named. you pay in context on every turn, plus a trust decision.yes
agents and productsclaude-powered software from crowdstrike, cursor, harvey, legora, lovable, snowflake (anthropic, sep 23 2026)a portion of your committed anthropic spendno
service partnersconsultants and integrators from the claude partner network, such as accenture, boston consulting group, deloitte (anthropic, sep 23 2026)a portion of your committed anthropic spendno

what the middle layer actually solves, and why it is not your problem

the agents and products layer is a procurement instrument, not a capability release. it lets a company that already signed a large anthropic commitment spend part of that budget on someone else's claude-powered software, instead of running a fresh vendor approval. snowflake, quoted in anthropic's own announcement, names the benefit this way:

"cutting procurement cycles and accelerating time to value"
snowflake, in anthropic's claude marketplace announcement, sep 23 2026

read that again as a solo operator: the thing being sold is the removal of a purchasing department. you do not have one. the friction it removes is friction you never had, which is exactly why this layer reads exciting and changes nothing for you.

the layer you can use today, and how it works

connectors and plugins are the layer that lands on your machine, and you do not need a spend agreement to reach them. anthropic's claude code documentation says claude code adds the official anthropic marketplace, claude-plugins-official, automatically the first time you start it interactively, and that you install from it with a slash command such as /plugin install github@claude-plugins-official (anthropic claude code docs, sep 2026). there is also a community marketplace whose plugins have passed anthropic's automated validation and safety screening, with each plugin pinned to a specific commit (anthropic claude code docs, sep 2026).

if the words connector, plugin, skill, and mcp still blur together, that distinction is the whole game and i mapped it in prompts vs skills vs plugins vs mcps. the setup order for connectors specifically is in claude connectors, explained.

the cost the announcement does not mention

every plugin you install charges you context on every single turn, whether you used it that turn or not. anthropic's documentation is unusually direct about this: the plugin details pane shows a context cost estimate, described as how many tokens the plugin will add to your context window every turn (anthropic claude code docs, sep 2026). a catalogue with 2,000 options and a one-click install is a very efficient way to fill your context with things you are not using.

claude code will even tell you when you have done it. it lists plugins you installed but have not used in at least two weeks, across at least ten sessions, under a "not used recently" header, with a last used line on each one (anthropic claude code docs, sep 2026). that feature exists because the failure mode is common. more on why a fuller context makes claude worse rather than better in more context is not better.

five checks before you install anything

the plugin details pane gives you everything you need to make this decision in about twenty seconds. run these five in order:

  1. context cost. what it adds to every turn, forever, not just the turns you use it. this is the number most people never look at.
  2. last updated. an abandoned plugin wrapping a changed api fails quietly, which is the worst way for a tool to fail.
  3. the will install list. the pane itemises the commands, agents, skills, hooks, and mcp and lsp servers arriving with it. read it before it arrives, not after.
  4. scope. user scope follows you across projects, project scope hits every collaborator on the repository, local scope stays in one repository. pick the narrowest one that does the job.
  5. trust, which is yours alone. anthropic states plainly that it does not control what mcp servers, files, or other software are included in plugins and cannot verify that they work as intended (anthropic claude code docs, sep 2026). a marketplace listing is a catalogue entry, not a warranty.

install or build: the question the marketplace actually forces

a 2,000-item catalogue makes installing feel like the obvious default, and for connectors to services you already pay for, it is. a notion or slack connector is plumbing, and writing your own plumbing is a bad use of a tuesday. the line moves when the plugin encodes judgment rather than access: how your offers are priced, how you qualify a lead, how your proposals get structured. that is the part no catalogue has, because it is yours.

i took this apart in more detail in prebuilt ai skills vs the ones you write yourself, and the anatomy of a plugin is in claude plugins, explained. the short rule: install access, write judgment.

if you want to be listed in it

anthropic says companies that build tools, products, or services for claude customers can be listed, and the path into the plugin layer is to build a connector or plugin using the model context protocol and agent skills (anthropic, sep 23 2026). worth knowing before you aim at it: the official marketplace is curated by anthropic and inclusion is at its discretion, the in-app submission forms add plugins to the community marketplace rather than the official one, and you can skip the queue entirely by creating and sharing your own marketplace (anthropic claude code docs, sep 2026).

the takeaway

three layers, one door. the connectors and plugins layer is open to you today and worth using deliberately, for access rather than for judgment. the agents and products layer and the service partner layer are procurement tools for companies with a committed spend agreement and a purchasing process, and reading them as a shopping list is how a week disappears. before you install anything from the open layer, look at the context cost, because the marketplace makes it very easy to pay it 2,000 times over.

the skills a catalogue cannot sell you

the ai builder toolkit is the set of claude skills that encode the judgment part: how offers get priced, how leads get qualified, how the work actually ships. built for a one-person business, installable today, no committed spend agreement required.

see the toolkit

newer to this and want the version with people in the room? the ai builders lounge is where the weekly builds happen.

or just follow along. new field notes most weeks on x, instagram, and tiktok.

one email when the next field note drops.

no course pitch, no daily emails. the notes, when they exist.

written by elisabeth hitz, certified in anthropic's ai fluency program (framework & foundations, and ai capabilities & limitations), plus claude 101 and claude cowork. primary source: "Claude Marketplace: plugins and connectors, products and agents, and service partners," anthropic, published september 23, 2026 (claude.com/blog/claude-marketplace). the three-layer structure, the 2,000+ figure, the named partners, the committed-spend terms, and the snowflake quote are from that announcement. install mechanics, the context cost estimate, the community marketplace screening, the "not used recently" behaviour, and anthropic's plugin trust disclaimer are from anthropic's claude code documentation, read september 24, 2026 (code.claude.com/docs/en/discover-plugins). the three-layer read, the install-access-write-judgment rule, and the five pre-install checks are mine, from setting claude up on real businesses. part of the verification layer series.