Automattic launched Spacefast on Monday, November 9, as a hosting and deployment service designed to let AI assistants publish websites and web apps directly from a chat or coding environment. The product works with ChatGPT, Claude, Cursor, Codex and other tools through plugins, a remote Model Context Protocol server, a command-line interface and an API.
Spacefast is open to the public now. Its central promise is narrower than an all-in-one AI app builder: users keep the model or coding agent they already use, while Spacefast supplies the deployment target, version history, domains, database, file storage and sharing controls. Automattic is positioning it as the infrastructure layer between an AI-generated project and a live URL.
That distinction matters because several features often treated as prerequisites for a production app are not generally available yet. Built-in visitor authentication and transactional email are in private beta, while Automattic’s launch announcement lists app authentication and broader team knowledge features as coming later. Spacefast can put a prototype, static site or supported web app online quickly today, but teams still need to check exactly which backend services their project requires.
What Spacefast can deploy now
Spacefast supports static HTML and applications built through documented paths for Next.js, Astro, TanStack Start and PHP. A new version is usually live within 10 seconds after the upload finishes, according to the product’s FAQ. Each publish creates a permanent version URL, while the main project address continues to point to the current release.
Every “Space” can include a MySQL-compatible database, file uploads, traffic statistics and comments attached to areas of a page. Custom domains are included across the public plans. New projects are private by default, with access available through invitations, passwords, revocable viewer links or a public setting; search indexing is controlled separately.
Users do not have to create an account before a first publish. An anonymous project receives a live address and a private claim link, but it normally expires 33 hours and 20 minutes after creation or the latest file-changing publish, with a 30-day cap from creation. Claiming the project removes that anonymous-expiry rule.
The service also separates AI costs from infrastructure charges. Spacefast does not sell model tokens or require users to change assistants. The free Personal plan includes $1 of monthly infrastructure usage, up to 20 Spaces and 20 custom domains. Go costs $5 a month and includes $5 of usage, 100 Spaces, hourly database backups and a choice of data center. Pro costs $15 a month, includes $15 of usage, supports up to five members without an additional seat fee and raises the limit to 500 Spaces.
The MCP server is the product’s control surface
Spacefast exposes five tools through its remote MCP server. publish uploads a project and returns live and fixed-version URLs; operation_status tracks a build; show_space reads deployments, domains, traffic, source files and logs; execute handles management work such as access and DNS; and resume_execution continues an operation after an approval.
This gives an assistant a relatively small, inspectable interface instead of broad access to a hosting dashboard. It also lets the same deployment live outside any one model’s conversation history. A user can begin in Claude, move to Cursor and later manage the project from another MCP-compatible client without changing its public address.
Local publishing excludes .env, .git, node_modules, .ssh and key files by default. Connections use OAuth, so the AI application does not receive a Spacefast API token that could be left in a chat transcript.
Agent permissions have useful limits
Spacefast divides access into read-only and agent modes. Read-only connections can inspect projects, versions, traffic and build logs. Agent access can publish new projects and create teams, but sensitive changes pause for human approval.
Those approval-gated actions include publishing over a live site, rolling back a release, buying a domain and changing DNS. Spacefast says approval requests wait for roughly 10 minutes. Connected agents cannot delete sites, change billing, manage team members or create credentials through this interface.
There is one deliberate exception: a brand-new publish can proceed without an additional confirmation. That makes the first prototype fast, but organizations should still decide whether an assistant may create externally reachable projects before enabling agent access. “Private by default” controls who can view a project; it does not replace code review, dependency scanning or testing of the application itself.
What is still beta or missing
The launch materials draw a line between infrastructure that is available now and application services that are not yet broadly released. Visitor sign-in and transactional email are marked private beta in the MCP documentation. Automattic also lists shared team knowledge, including brand and playbook files, among the features still on the way.
That means a public marketing site, prototype, game or internal review build fits the current product more cleanly than an app whose core workflow depends on user accounts, password recovery, receipts or invitation emails. Developers can bring their own implementations, but doing so changes the security and maintenance burden that Spacefast is meant to reduce.
Portability is also uneven. Users can download files from retained versions, and static sites can generally move to another host without modification. Spacefast’s documentation warns that applications built around its runtimes may require code changes when moved. Teams evaluating the platform should identify which database, storage, authentication and runtime interfaces are proprietary before committing a production workload.
How Spacefast compares with AI app builders
Products such as Lovable, Emergent and other prompt-to-app services combine code generation with a managed deployment path. Spacefast instead starts from the assumption that the user already has an assistant or coding agent. It competes for the publishing and operations layer rather than requiring the model experience to move inside a new builder.
That approach gives Automattic a way into AI-built software without turning WordPress into the universal answer for every project. Spacefast founder and CEO Batuhan İçöz told TechCrunch that some sites will not need a content-management system. The service runs on the infrastructure behind WordPress.com and Tumblr, which Automattic says serves more than 15 billion page views per month.
Spacefast began as an internal project during Automattic’s Radical Speed Month, initially as a static host for projects created with AI. Demand for a backend pushed it toward databases and managed application services. It soft-launched in September before becoming a separate Automattic company under İçöz.
What teams should verify before deploying
For a low-risk prototype, the evaluation is straightforward: confirm the supported framework, claim the project before the anonymous link expires, set the intended viewer permissions and test rollback. A production review needs more.
- Authentication: determine whether the private-beta visitor sign-in feature is available or whether the app must supply its own identity system.
- Email: do not assume receipts, invitations or passwordless links are ready; transactional email is also in private beta.
- Secrets: verify that no credentials are embedded in generated source code, client-side bundles, prompts or uploaded files.
- Data location and backups: paid plans can choose a data center, while backup frequency and retention vary by plan.
- Spending controls: Go and Pro support a cap, starting at twice the plan price, with alerts at 50%, 75% and 100%.
- Exit path: download a retained version and document any Spacefast-specific runtime or database dependencies before launch.
- Agent access: give assistants the smallest team scope and permission level needed, and test the global agent-access cutoff.
Spacefast arrives with a credible infrastructure base and a more thoughtful agent-permission model than a simple “deploy” button suggests. Its immediate strength is turning AI-made projects into shareable, versioned deployments without forcing users into another coding interface. Whether it becomes a production platform will depend on how quickly the beta application services mature, how portable stateful apps prove to be and whether teams keep normal software review in the loop after publishing becomes almost frictionless.
