I build most of my sites by talking to an agent. Claude Code writes the components, I review the diff, we ship. The code part is fun now. The part that still hurts is the CMS.
The question is never just "can I build it." It's "what happens after." Who fixes a typo when the client wants it done at 9pm? Can the agent that wrote the site also write the blog posts? Will I still be paying for plugins in two years?
So this is a buyer's roundup, not a pitch. I run Slim Minima, one of the twelve options below, and I'll tell you where it loses.
If you write code with an AI agent and hand the finished site to someone non-technical, the CMS that fits is the one your agent can read and write through an API, and the one your client can edit without you. Most popular CMSes are good at one of those, not both.
What a vibe coder actually needs from a CMS
"Vibe coding" gets mocked, but the working style is real: you direct an AI agent, it produces code, you stay in the loop on architecture and review. If that's you, your CMS requirements differ from a classic web shop.
Here's the checklist I'd hold any CMS to:
- Code-first, not builder-first. You want to write components, not drag boxes. The content tool should fit around your code.
- An API the agent can use. If your agent can't read and write content programmatically, you're back to pasting into a web form. A REST or GraphQL API is the floor.
- MCP support, or something close. The Model Context Protocol lets an agent like Claude or ChatGPT talk to a tool directly, so your agent can publish a post without you in the middle.
- A clean schema. You should be able to read the content model in a few minutes. Schema sprawl is where projects rot.
- No plugin lock-in. Plugins are convenient until one breaks your update, holds your data, or stops being maintained. Code you own does not abandon you.
- Easy handoff to a non-technical owner. A custom site with no admin panel sends every tiny edit back to you, forever.
- Cheap hosting. Side projects and small clients should not need a monthly server bill to stay online.
No tool nails all seven. The trade-offs are the whole story. A quick word on stack, since vibe coders ask it first: most AI agents are strongest in TypeScript, React, and Next.js, so a CMS built on that stack is usually easier to direct an agent through than one built on PHP or a closed visual canvas.
The short verdict
If you want one line: for a vibe coder building marketing and content sites that get handed to a non-technical owner, the strongest fits in 2026 are Slim Minima (handoff plus free hosting plus a built-in agent API and MCP) and Payload (Next.js-native with serious data modeling, if you don't mind running it). TinaCMS and Decap are great if you want content as Markdown files in your repo. Directus has the best agent story over a SQL database. Sanity is excellent for structured, multi-channel content. Webflow and Framer win on visual polish but fight the code-first workflow. WordPress has the deepest ecosystem and the most baggage. Strapi, Keystone, and Contentful each fill a clear niche below.
Now the honest detail, one tool at a time.
WordPress
WordPress runs a large share of the web, and that is not an accident. The ecosystem is enormous. Whatever you need, a plugin or theme probably exists, and a freelancer who knows it is easy to find.
Platform. A PHP monolith with a MySQL database. Not React, not Next.js, which already puts it at a distance from most AI-assisted workflows.
Strengths. Unmatched plugin and theme catalog. A non-technical owner can usually figure out the editor. Hosting is cheap and everywhere. The WordPress REST API is mature, so an agent can read and write posts.
Weaknesses. The plugin model is also the problem. Updates break each other, security holes come through third-party plugins, and the data model gets buried under meta fields. There's no official MCP server.
Can you vibe code with it? Partly. An agent can drive the REST API, but you're working in PHP and a theme system most agents handle less fluently than a React codebase. More on that in why WordPress does not work well in the AI age.
Payload CMS
Payload is a TypeScript-native, code-first CMS. As of version 3 it installs directly into a Next.js app, so the admin panel and your site live in the same codebase.
Platform. Next.js-native, React admin, Node backend, Postgres or MongoDB. MIT licensed and self-hosted.
Strengths. The data modeling is genuinely strong: relationships, access control, hooks, and a generated admin panel that stays in sync with your typed config. You get REST, GraphQL, and a local Node API out of the box. It now ships an official MCP plugin, so an agent can read and write content directly while respecting your access controls.
Weaknesses. You run it. That means a database, a Node server, and the ops that come with both. There's more upfront setup than a hosted tool.
Can you vibe code with it? Yes, and well. The whole config is TypeScript an agent can scaffold and edit, and the MCP plugin closes the loop. The cost is that you own the infrastructure.
TinaCMS
Tina is a Git-backed CMS with a visual editor. Your content lives as Markdown or JSON files inside your repository, not in a database.
Platform. TypeScript, built on React with Next.js as the first-class framework. Apache-2.0 and free to self-host, with an optional hosted backend (TinaCloud).
Strengths. Maximally code-first: content is plain Markdown files an agent can read, write, and commit like any other code. No database to run if you self-host. The visual editor lets a non-technical owner edit on the page.
Weaknesses. It's tied tightly to Next.js and React, the API is GraphQL only, and real collaboration is paywalled (the free tier caps at two users). There's no official MCP server yet, only a prototype.
Can you vibe code with it? Yes, naturally, because the content is just files. An agent edits a .md file and commits it. The handoff to a non-technical client is the weaker side.
Sanity
Sanity treats content as structured data, not pages of HTML. You define schemas, and content lives as portable, queryable documents. Its real-time collaboration and editing experience are among the best in the category.
Platform. A hosted content platform. The studio (the editing app) is React, but the content itself is API-first, so the website is yours to build in any framework.
Strengths. Structured content done properly, which pays off when the same content feeds a website, an app, and a newsletter. Real-time multiplayer editing. A flexible query language (GROQ) plus a solid API make programmatic access friendly. The Sanity docs are thorough.
Weaknesses. It's API-first, so the "website" part is yours to build. Pricing is usage-based, so heavy traffic costs more as you grow. The structured-content mindset is more setup than a simple blog needs.
Can you vibe code with it? Yes for the data layer, since the API and GROQ are agent-friendly, and Sanity has been adding AI features. You still build and wire the front end yourself.
Strapi
Strapi is a popular open-source, self-hosted headless CMS. You model content in an admin UI or in code, and you get a REST and GraphQL API on top.
Platform. Node and TypeScript, React admin, your own database. Open source and self-hosted.
Strengths. You own your data. A friendly admin for building content types. A large plugin ecosystem if you want it, and good documentation with a big community.
Weaknesses. Like Payload, you run the server and the database. The plugin ecosystem reintroduces some of the lock-in risk you were trying to avoid. No official MCP server, so agent access is the REST route you wire up.
Can you vibe code with it? Yes through its API, but the content modeling is admin-driven rather than config-as-code, so an agent does less of the structural work for you.
Directus
Directus wraps any SQL database with an instant API and a polished admin app. Point it at a Postgres or MySQL database and it generates the rest.
Platform. Node and TypeScript backend, with a Vue admin (not React). Source-available rather than strictly open source as of its recent license change, and self-hostable.
Strengths. The best agent story on this list after the built-in ones: Directus ships an official MCP server, plus instant REST and GraphQL over whatever SQL database you already have, and the admin hands off cleanly to clients.
Weaknesses. It's database-first, not git or file based, so content is not in your repo. Self-hosting needs a Node server and a database. The license is now source-available, not OSI open source, which some teams care about.
Can you vibe code with it? Yes. The official MCP plus generated APIs mean an agent can model and manage content directly. The setup is heavier than a static, file-based tool.
KeystoneJS
Keystone is a developer-first CMS where you define your schema in TypeScript and it generates a GraphQL API and an admin UI.
Platform. TypeScript end to end, React admin, Prisma and a SQL database, Node runtime. MIT licensed and fully self-hosted, with no first-party hosted option.
Strengths. Schema-as-code an agent can reason about from a single config file, with real type safety. Free and ownable. Strong if your content is relational and you like defining it in code.
Weaknesses. No official MCP or AI tooling, GraphQL only, and no hosted option, so you own the database, deploy, and ops. Development moves at a steady but modest pace.
Can you vibe code with it? Yes for the schema, since it's pure TypeScript, but agent access to content is plain GraphQL you wire yourself, and you carry all the hosting.
Decap CMS
Decap, formerly Netlify CMS, is a Git-based admin for static sites. Content is files in your repo, edited through a simple web UI.
Platform. A React admin you embed in your static site, configured by one file. MIT licensed, free, and self-hosted. Not built on Next.js, but it works alongside any static generator.
Strengths. Zero infrastructure and zero cost. Content is just files you fully own, and it drops onto any static site. Hard to beat for a simple, free, ownable blog.
Weaknesses. No content API or backend of its own, no native AI or MCP, and a dated admin. Agents can't talk to it directly; they edit the underlying files instead.
Can you vibe code with it? Yes in the file sense: an agent edits Markdown and commits. There's no programmatic content API to drive, so anything beyond files-in-a-repo is on you.
Contentful
Contentful is a mature, hosted headless CMS aimed at larger teams. Content-as-a-service with a clean editing UI.
Platform. Proprietary cloud SaaS, no self-host. It pairs with React and Next.js front ends but is not built on them.
Strengths. Well-documented REST and GraphQL APIs, a polished editor, and a mature official MCP server, so it's reliably agent-writable.
Weaknesses. A steep cost cliff: the free tier jumps straight to a plan around 300 USD per month with little in between, per the Contentful pricing page. Not code-first or git-based, and there's no self-host escape hatch.
Can you vibe code with it? Yes through the API and MCP, but you're renting a platform, and the price step makes it a poor fit for small client sites.
Webflow
Webflow is a visual-first site builder with a real CMS attached. The design control is excellent, and the output is clean.
Platform. A hosted visual canvas. You design in the browser, not in code, so it sits furthest from a code-first workflow.
Strengths. The best visual design experience on this list. You can build pixel-precise layouts without writing CSS, and clients can edit content in a clear editor. Hosting is included, and there's an official API for content.
Weaknesses. It fights the code-first workflow. You design in Webflow's canvas, so an agent that writes components has little to do. You don't own the platform or hosting the way you own code, and exporting is limited.
Can you vibe code with it? Not really. The point of Webflow is visual building, not directing an agent over a codebase. It's a strong tool aimed at a different person.
Framer
Framer is a visual website builder with a built-in CMS, in the same family as Webflow. It began as a design tool and grew into a full site builder, and its CMS (Collections) handles blogs and other dynamic content.
Platform. A hosted visual canvas, SaaS only. Sites run on Framer's own infrastructure, with no HTML export and no self-hosting.
Strengths. Fast, polished visual design with a real CMS behind it. Clients can edit content in a clear interface, hosting is handled, and there's an official Server API that can read and write CMS collections programmatically. Free to start, with paid site plans from 10 USD per month, per the Framer pricing page in June 2026.
Weaknesses. It's visual-first and closed. Framer sites run on React under the hood, and you can drop in custom React code components, but code is an escape hatch, not the build surface, and you cannot export or own the result. There's no official MCP server, only a community plugin.
Can you vibe code with it? Not really. Like Webflow, Framer is built for designing on a canvas, not for directing an agent over a codebase you own. An agent can poke at the Server API, but it cannot build the site as source.
Slim Minima
Slim Minima is the CMS I build. It's an open-source (MIT) content system on Next.js 16, React 19, Drizzle, and Neon Postgres, designed for exactly this workflow: vibe code the site, hand off a CMS. So treat this section as the most biased one.
Platform. Next.js and React, code-first, self-hosted on infrastructure you own.
Strengths. It's built around handoff. You write the components in code, and your client gets an admin panel to edit content without touching you. It runs on a free stack: Vercel Hobby, Neon free, and Cloudinary free cover roughly 10,000 to 30,000 monthly visitors at no cost beyond a domain, per Vercel, Neon, and Cloudinary pricing.
There's a REST API and a built-in MCP server, so an agent like Claude or ChatGPT can create and update posts, upload images, and add categories directly. There are no plugins: you extend the site with TypeScript and React, so there's no plugin update to break and no marketplace to depend on. The schema is small enough to read in one sitting.
Weaknesses. "No plugins" is a real trade-off. If you want a feature, you write code or have your agent write it; there is no one-click install. The MCP server intentionally has no delete tool. It's young, at v0.1.3, and largely a solo and community project, so the bus factor is real.
The honest mitigation is the MIT license: you keep the code and the database forever and can self-maintain or fork. It's built for content sites, not real-time or write-heavy apps.
Can you vibe code with it? Yes, by design. It's the same Next.js and React an agent is already strong in, the schema is tiny, and the MCP server lets your agent write content with no glue code. The longer story is in why I built Slim Minima, and the security reasoning in is Slim Minima secure.
The comparison table
Same attributes across all twelve, kept short for mobile:
| CMS | Built on | Code-first | Agent API | Official MCP | Database | Self-host |
|---|---|---|---|---|---|---|
| WordPress | PHP | No | REST | No | MySQL | Yes |
| Payload | Next.js | Yes | REST + GraphQL | Yes | Postgres / Mongo | Yes |
| TinaCMS | React / Next | Yes (files) | GraphQL | No | None, files in git | Yes |
| Sanity | API-first | Partly | REST + GROQ | No | Hosted | Hosted |
| Strapi | Node | Yes | REST + GraphQL | No | SQL | Yes |
| Directus | Node / Vue | Partly | REST + GraphQL | Yes | Any SQL | Yes |
| KeystoneJS | Node / React | Yes | GraphQL | No | SQL | Yes |
| Decap | React | Yes (files) | Git only | No | None, files in git | Yes |
| Contentful | Hosted SaaS | No | REST + GraphQL | Yes | Hosted | No |
| Webflow | Visual | No | REST | No | Hosted | No |
| Framer | Visual (React) | No | Server API | No | Hosted | No |
| Slim Minima | Next.js | Yes | REST | Yes | Postgres | Yes |
A note on fairness: the MCP column tracks an official, built-in server as of June 2026. Several vendors are shipping these quickly, and any tool with a REST or GraphQL API can be driven by an agent if you wire it up. So "No" means the integration is yours to build, not that an agent cannot use the tool.
Limitations of this roundup
This is a snapshot in June 2026, and these tools move fast. Pricing changes, licenses change (Directus moved to a source-available license recently), and several tools may ship or expand MCP servers after this is published. Verify the current state on each tool's own site before you commit, especially pricing, licensing, and API limits.
I also have a stake here. I tried to credit each tool's real strengths and state Slim Minima's real weaknesses, but weight my section accordingly.
Who should skip the "vibe coder" framing entirely
If you think in canvases, not components, Webflow or Framer will make you happier than any code-first tool. If your team is already fluent in WordPress with a maintenance routine that works, switching costs more than it saves. And if you're building a real-time, write-heavy app rather than a content site, none of these CMSes are the core of your stack.
Frequently asked questions
What is the best CMS for vibe coders in 2026? There's no single answer, but for AI-assisted developers building content and marketing sites that get handed off, the strongest fits are Slim Minima (free stack, built-in MCP and API, code-first) and Payload (Next.js-native with strong data modeling if you can host it). Tina and Decap are best if you want content as Markdown in your repo, Directus has the best agent story over a SQL database, and Webflow is best for design-led work.
Which CMS works best with Claude Code or an AI agent? Any CMS with a REST or GraphQL API can be driven by an agent. The smoothest setup is one with a built-in MCP server, since the agent connects directly. As of June 2026, Slim Minima, Payload, Directus, and Contentful ship official MCP servers. With WordPress, Tina, Sanity, Strapi, Keystone, Decap, or Webflow you wire the agent to their API or files yourself.
Do I need a headless CMS for AI-assisted development? Not necessarily. Headless tools like Sanity, Strapi, and Contentful give you an API but leave the front end to you. A coupled-but-code-first CMS like Slim Minima or Payload gives you the API plus a built-in renderer and admin panel, which is less to assemble. Pick based on how many channels the content has to feed.
Which CMSes are built on Next.js or React? Payload is Next.js-native. Slim Minima runs on Next.js and React. Tina, Strapi, Keystone, Decap, and Sanity's studio use React in various ways. WordPress (PHP) is not React or Next.js. Webflow and Framer are visual builders: Framer renders React under the hood, but you build on a canvas rather than in code, which is part of why both fit an agent workflow less well.
How much does it cost to run a CMS for a small site? It varies widely. WordPress needs cheap hosting. Payload, Strapi, Keystone, and Directus need a server and database you pay for. Sanity is usage-based, Contentful jumps to around 300 USD per month after its free tier, and Webflow is plan-based. Decap is free, and Slim Minima runs free up to roughly 10,000 to 30,000 monthly visitors on Vercel, Neon, and Cloudinary free tiers, with the domain as the only required cost.
Related reading
- Why I built Slim Minima
- Why WordPress does not work well in the AI age
- Is Slim Minima secure?
- Is Slim Minima a good CMS for web design agencies?
- How far does the free stack stretch?
My verdict
Match the tool to your job. If you're a vibe coder shipping marketing sites or blogs that a client will edit, and you want a free stack with a built-in agent API and no plugins, Slim Minima fits, with the honest caveats that it's young and you extend it with code.
If you need rich relational data and can run your own infrastructure, choose Payload. If you want content as files in your repo, choose Tina or Decap. If you want an agent talking to a SQL database, choose Directus. If your content feeds many channels, choose Sanity. If design control is the whole point, choose Webflow or Framer.
If you live in the WordPress ecosystem already, stay there and keep up with the updates. There's no best for everyone, only the best for how you work.