Home › Rules › Lovable MCP + Claude Code

Lovable MCP server + Claude Code: setup, plugin and skill

Commands and details from Lovable's and Anthropic's own docs, checked Oct 28, 2026. Sources below.

Short answer: Lovable runs an MCP server at https://mcp.lovable.dev. Once Claude Code is connected to it, you can ask Claude Code, in plain English, to create Lovable projects, send changes to Lovable, read the code, check what changed, and publish. Lovable says it works on all plans, including Free. There are three pieces, and you only need the first: the MCP server (one command), the Lovable plugin (the same server plus a few ready-made commands), and the skill file (a how-to sheet that teaches Claude Code to use the server well). These calls run on your real Lovable account and use real credits.

MCP server, plugin or skill: which one do you need?

PieceWhat it is and how you add itUse it if
MCP serverThe connection itself. MCP (Model Context Protocol) is an open standard that lets an AI tool find and use outside tools. Lovable's server gives Claude Code a set of Lovable tools. You add it with one command.You want the plain connection and will ask for things in your own words
Lovable pluginA Claude Code plugin made by Lovable, listed in Anthropic's official plugin marketplace. It bundles the same MCP server and adds commands for building, changing and database work, which ask you before spending credits or publishing. You add it with /plugin install.You'd like ready-made commands and safety prompts instead of writing your own
Skill fileA markdown file Lovable publishes (skill.md). It lists every tool and the order to use them in, so Claude Code knows how to drive Lovable without extra prompting. You save it in your skills folder.You want Claude Code to make fewer wrong turns with the Lovable tools

Lovable's docs give the command and the plugin as two ways to do the same job, so pick one. The skill file works with either.

Don't mix these up with three other Lovable features that have similar names:

What you need first

Option 1: add the MCP server (one command)

  1. Open a terminal in your project folder and run this command. It's the exact line from Lovable's docs:
Terminal: add the Lovable MCP server
claude mcp add --transport http lovable "https://mcp.lovable.dev"
  1. Decide where it should work. By default, Claude Code adds a server to "local" scope, which means only you, and only in the project folder you ran it in. To have Lovable in every project on your computer, add --scope user instead:
Terminal: same server, for all your projects
claude mcp add --transport http --scope user lovable "https://mcp.lovable.dev"
  1. Start Claude Code with claude. The first time a Lovable tool runs, Claude Code opens a browser window. Log in to Lovable and approve the access it asks for.
  2. Check that it worked. Lovable's docs give two checks: type /mcp and look for the lovable server, or just ask Claude Code to list your Lovable workspaces.
Inside Claude Code: check the connection
/mcp
list my Lovable workspaces

In a normal terminal, claude mcp list shows each server's status, such as Connected or Needs authentication.

Option 2: install the Lovable plugin

The plugin connects the same server for you and adds commands. Lovable's docs give this line, typed inside a Claude Code session:

Inside Claude Code: install the Lovable plugin
/plugin install lovable@claude-plugins-official
  1. Review the details. Claude Code's docs say this opens the plugin's details first, so you can see what it adds before it installs.
  2. Pick a scope. "Install for you" puts it in every project on this computer; the other two options limit it to the current repository, either for everyone working on it or only you.
  3. Read the last line of the summary. If it says Run /reload-plugins to apply, Claude Code does that for you when the panel closes.
  4. Find the commands. Type / and look for entries under the plugin's name, in the form /lovable:.... The plugin's open-source repository has three: build (make a new app from a description and optionally publish it), iterate (send a change and show the diff), and db (set up or query the project's database).

Prefer your normal terminal? Claude Code's docs give the shell form too: claude plugin install lovable@claude-plugins-official. Anthropic's official marketplace is added automatically the first time you open an interactive Claude Code session.

Add the Lovable skill file (optional, works with either option)

A skill in Claude Code is a folder with a SKILL.md file of instructions that Claude loads when the task calls for it. Its docs say to download it into your client's skills folder, for example .claude/skills/lovable-mcp/SKILL.md for Claude Code. On a Mac or Linux, from your project folder:

Terminal: save Lovable's skill file into this project
mkdir -p .claude/skills/lovable-mcp
curl -o .claude/skills/lovable-mcp/SKILL.md https://mcp.lovable.dev/skill.md

What Claude Code can do with your Lovable projects

Lovable's docs list these tool groups. Claude Code picks the right tool from what you ask.

Example prompts from Lovable's docs: "Show me what changed in the last three edits to my project" and "Deploy my project [ID] and give me the live URL".

Matching kit: MVP Spec-to-Build Pack ($7)

The free rules set the boundaries. The MVP pack gives you the build order: a 1-page spec to fill in first, then 7 prompts in sequence (skeleton, review, login, database, security, one feature at a time, polish), plus a "did I leak user data?" checklist.

See the MVP Spec-to-Build Pack ($7) →

Instant download. 7-day refund, no questions asked.

Credits, publishing and your data: read this before your first prompt

Troubleshooting checklist

Add a Lovable section to your CLAUDE.md or AGENTS.md

The plugin's commands ask before spending credits or publishing. If you use the plain MCP server, write those rules down once in your project's instructions file so Claude Code follows them every session. Claude Code reads CLAUDE.md at the start of each session, and reads AGENTS.md when there's no CLAUDE.md. Paste this section into whichever one you use and fill in the brackets:

CLAUDE.md or AGENTS.md: Lovable section
## Lovable (through the Lovable MCP server)
- My Lovable project: [project name]. Project ID: [paste it once you know it].
- Ask me before any call that uses Lovable credits (creating a project or sending Lovable a message). Tell me what you plan to send first.
- For anything bigger than a small change, send it to Lovable in plan mode first and show me the plan.
- After Lovable finishes a change, show me the diff and tell me how to check it in the preview.
- Never publish (deploy) without my OK. Tell me who will be able to see the live link.
- Before running SQL on the project's database, show me the exact query. Never change or delete data, or the table structure, without my OK.
- Don't change Project Knowledge or workspace skills unless I ask. Those updates replace the old text completely.
- Work on one Lovable project at a time, and only the one I named.

Anthropic suggests keeping a CLAUDE.md under 200 lines, and Claude treats it as guidance, not a hard lock. If the project is also synced to GitHub, Lovable's agent reads AGENTS.md and CLAUDE.md files in the repo too, so keep this section short. Rules for Lovable itself belong in Project settings → Instructions (our Project Knowledge examples cover that). More starters are in CLAUDE.md examples and AGENTS.md vs CLAUDE.md.

MCP server or GitHub sync?

With the MCP server, Lovable still writes the code and Claude Code works like a remote control. With GitHub sync, the code comes to your computer and Claude Code edits the files itself. To keep building in Lovable, use the MCP server. To edit the code directly, see our Lovable to Cursor or Claude Code guide, which comes with a ready AGENTS.md. For a side-by-side of the two tools with prices, see Lovable vs Claude Code.

Next step for Lovable and Claude Code: Agent Skills

Skills are folders your AI coding agent loads only when the job comes up, and Claude Code, Cursor and OpenAI Codex all read the same SKILL.md format. The Vibe Coder Skills Pack ($9) is 6 ready-to-copy SKILL.md skills (fix-one-bug, stop-the-loop, leak-check, ship-check, explain-project, builder-handoff) for Claude Code, Cursor and OpenAI Codex. Also included in the 6-kit bundle.

Get the Skills Pack for $9 on Gumroad → See what's inside

One .zip of skill folders plus one markdown file. 7-day refund, no questions asked.

Best value

All 6 kits for $29 — or $20.30 with code WELCOME30 (through Nov 30)

Connecting the tools is the easy part. The kits tell your AI what to build, step by step: a small game, a Discord bot, an email-triage agent, an MVP, or a cleanup of messy code.

Get all 6 kits for $29 → See what's in each kit

$29 instead of $48 for the six kits bought separately; WELCOME30 takes 30% off that at checkout ($20.30). Instant download from Gumroad. 7-day refund, no questions asked.

Related

Sources

Commands, tool names and plan details checked Oct 28, 2026 on Lovable's and Anthropic's own docs and pricing pages. These tools change often. If something here no longer matches, tell us at [email protected].

Get a free starter kit: the Discord Bot Builder ($6)

  • The Discord Bot Builder kit, free (it's $6 on Gumroad): from the Discord Developer Portal to a running bot, an agent brief (AGENTS.md / CLAUDE.md) with token-safety rules, a starter bot with /roll and /ping, and six feature prompts. Emailed with your download link.
  • Bonus: the whole rules library in one zip: 137 ready-to-use rules files (AGENTS.md, CLAUDE.md, Cursor .mdc rules, Lovable/Bolt Project Knowledge) for 30 projects.

After that, an occasional email when we publish a new guide, rules file or kit.

Free. No spam. Unsubscribe anytime. See our privacy note.