Lovable Feature Roundup: What actually mattered in May and June 2026

Why this roundup

We usually write a dedicated post for almost every Lovable release – Subagents, Skills, the mobile app, Lovable Cloud vs. Supabase. The last six weeks shipped so much in parallel that writing each up individually would have buried the signal. So: one roundup instead.

Sorted by "what changes your workflow tomorrow?", not by release date. Cross-checked against the official Lovable changelog and the aggregated MyAI Guide tracker.

Tier 1 – Actually changes how you work

1. Subagents (27 May)

The most important release of the window. Lovable can now spin up multiple read-only helper agents in parallel, each with its own context window. On large projects you feel it instantly: sharper answers, less "let me check again", and often a lower bill because the head agent no longer needs the entire repo crammed into its own context.

We did a full deep dive on Lovable Subagents including the kitchen analogy and when to invoke them explicitly.

2. Lovable MCP – any agent can now drive Lovable (17 June)

The strategically most important release of the window: Lovable now ships its own MCP server. That turns Lovable itself into a tool every MCP-capable client can call – Claude (Desktop & Cloud), Cursor, Codex, ChatGPT, your own agents. Reference projects, read code, propose changes, deploy – all without opening a Lovable tab.

This is exactly the pattern we described in No MCP, no Party: if you don't ship MCP, you're not in the user's workflow. Lovable ships – and positions itself as the default build tool inside any agent pipeline. Concretely for us: strategy session in Claude, research via other MCPs, commit via Lovable MCP – all in one thread.

It also reframes our earlier piece Claude Design vs. Lovable: the two worlds are growing together natively instead of cannibalising each other – but the bigger signal is that Lovable is now an MCP-first platform.

What this looks like in practice – Claude Desktop as the control room:

1. One-time setup: In Claude Desktop go to Settings → Connectors → Add custom connector and register the Lovable MCP server (https://mcp.lovable.dev/, auth via your Lovable account). Claude pulls the tool list automatically.
2. Reference the project: In chat, name the project via slash command or URL – @lovable list-projects shows all of them, @lovable open <project-id> makes one the active context.
3. Hand off the task: Instead of prompting yourself, give Claude an outcome:

"Read the Lovable project marketing-site. The Pricing page is missing a FAQ block. Pull the five most common pricing questions from our Notion sales notes (via the Notion MCP), build a shadcn FAQ accordion from them, commit it to a branch faq-pricing, and send me the preview URL."

Claude then orchestrates on its own: Notion MCP for research → Lovable MCP for the code change, commit and preview. You see every tool call in the chat, can interrupt anytime, and end up with a preview link – without ever opening the Lovable editor. That multi-MCP flow is the real jump: not "Claude instead of Lovable", but Claude as conductor, Lovable as the build instrument.

3. New Preview Toolbar replacing Visual Edits (15 June)

The old "Visual Edits" are gone. In their place sits a Preview Toolbar that lets you, directly on the live preview:

• pick elements and edit text inline,
• draw annotations,
• attach comments to specific spots that the agent receives as a structured prompt.

Sounds cosmetic, but this is the point where Lovable becomes genuinely multi-stakeholder-ready: your clients can comment in the preview without needing to understand the chat at all.

4. Publish & Deploy from chat (13–16 June)

Until now, publishing was a click in the editor UI. Now you can say "please deploy" in chat – Lovable verifies settings, runs security checks, schedules the deploy and ships after confirmation (or directly, if auto-approve is on). For CI-style workflows ("hotfix → ship") this is noticeably faster.

Important: it stays a confirmed step, not a silent auto-deploy. Exactly right.

5. Slow-query analysis for Lovable Cloud (15–16 June)

The most surprising release for us. Lovable Cloud now analyses slow PostgreSQL queries, ranks them by execution time, inspects the query plan and suggests targeted indexes. That's a feature level you'd expect from managed databases like Supabase Pro or Neon – and it lands at the free tier.

For anyone whose Lovable app has reached a few thousand users and suddenly feels "slow", this is the moment you no longer need external monitoring just to add your first index. More context: Lovable Cloud vs. Supabase.

Tier 2 – Quality-of-life you'll quickly stop wanting to live without

6. Code-line references in chat (17–18 June)

Next to every line number in the editor there's now a "+" chip that drops the line as an interactive pill into chat. Tiny on paper, transformative for bug reports: less "have a look at useAuth", more "these three lines are the problem." Subagents (see Tier 1) get even more efficient because of it.

7. Build with URL: now accepts any web page (19 June)

"Build with URL" used to accept images only. Now you can link to any public web page and Lovable uses its layout, content and styling as a reference. Quick win for pitch iterations ("build me this, but for our industry").

8. Publish button dot indicator (19 June)

The Publish button now shows a small dot whenever your project has changes that aren't live yet. Tiny detail that prevents you from sharing an old build with an investor.

9. Read-only project sharing (15 June)

For Business and Enterprise workspaces: share projects with "Can view" permission so teammates can open and reference them without accidentally editing code. The exact feature we'd been missing in multi-builder workspaces.

10. External domain transfers (15–17 June)

Transfer domains from other registrars (GoDaddy, Namecheap, IONOS …) directly into the Lovable workspace and manage renewals + DNS there. Saves the one quarterly login at that registrar you hate anyway.

Tier 3 – Strategic, (still) invisible day-to-day

• Google Cloud expansion (3 June): Lovable and Google Cloud deepened their partnership – more Gemini models, AI-optimised infrastructure, global rollout. Take-away: Lovable will keep scaling cheap-and-fast without locking into one model vendor.
• "Setting the standard for agentic development" (13 May): Joint paper with AIUC on safety standards for agentic coding. Reads like lobbying – which is exactly the point. Whoever wants enterprise buyers has to set standards before others do.
• Automatic security checks (1 June): Lovable enabled more default checks (RLS hints, secrets scanning, etc.). Invisible for most projects – until they prevent a finding that would otherwise have cost real money.
• TanStack Start support (May): Lovable can now build TanStack-Start apps. Niche, but interesting for anyone coming from React-Router land who wants more full-stack heft.
• LinkedIn Skills: If your Lovable usage profile qualifies, you can claim it as a LinkedIn skill. More personal branding than product, but a nice signal for the builder community.

Outlook: Lovable will bury the website-builder world (a thesis by Malte Lensch)

A thesis our managing director Malte Lensch has been repeating in every other strategy call for months – and the last few weeks of releases make it more plausible, not less:

Sooner or later, Lovable will specialise on websites in particular – and bury WordPress, Wix, Squarespace, Jimdo and every builder system that fails to pivot AI-first in time.

The evidence keeps stacking: Build with URL for arbitrary websites, external domain transfers, publish-from-chat, a Preview Toolbar built for non-techies, read-only sharing for stakeholders. This isn't only an app builder anymore – piece by piece it becomes the full website lifecycle in one tool, including domain, deploy, edits for non-coders and feedback loop.

Read alongside our end-of-WordPress-and-Webflow piece and the website-builders-vs.-AI-stack analysis, the pattern is clear: builders have no AI-first DNA – they bolt chatbots onto a drag-and-drop UI. Lovable comes from the other direction: AI as the editor, code as the output, hosting and domains as a commodity. The question isn't whether Wix & co. catch up, but whether they can – without a full architecture reset.

The take-away for our clients: anyone starting a new website project on a classic builder in 2026 is building for a world that won't exist in 24 months.

What we still miss

A few things Lovable should ship in the next wave:

1. Schema migration diffs for Lovable Cloud – slow-query analysis proves the DB tooling is serious. Logical next step: a "diff & rollback" UI for migrations.
2. Preview Toolbar in mobile browsers – currently desktop-shaped.
3. More granular subagent controls – "use at most 2 subagents for this task" as an explicit hint.

How we actually use this stack

At AI First Builders the setup has settled like this after six weeks:

• Subagents run implicitly on every non-trivial task; we only invoke them explicitly during mega-refactors.
• Preview Toolbar is now the #1 channel for client feedback while we're building.
• Lovable MCP is our new default entry point: strategy and research threads happen in Claude or Codex, commits land directly in the Lovable project via MCP.
• Publish-from-chat is on for marketing pages, off for cloud apps with real user data – there, publishing stays a conscious click.

Net-net: the past six weeks proved Lovable doesn't just ship "more features per sprint", the roadmap behind it is consistent. Subagents, MCP, Preview Toolbar, Cloud DB tools – they're the same story: moving from a single-agent code generator to a multi-agent platform for entire product teams.

---

Last updated: 21 June 2026. Cross-link: Lovable Feature-Roundup – die Mai/Juni-2026-Highlights.