mcp
Health Uyari
- License — License: Apache-2.0
- No description — Repository has no description
- Active repo — Last push 0 days ago
- Community trust — 10 GitHub stars
Code Gecti
- Code scan — Scanned 5 files during light audit, no dangerous patterns found
Permissions Gecti
- Permissions — No dangerous permissions requested
Bu listing icin henuz AI raporu yok.
Lovable plugin for Claude
Build, run and grow real apps from Claude. This plugin bundles the official Lovable MCP server with 13 skills and three commands, for Claude, Claude Code and Cowork.
Claude does the thinking in the conversation. Lovable builds the app, hosts it, gives it a database and a URL. The skills in this plugin connect the two so an idea, a spreadsheet or a Claude Design ends up as something that exists and keeps running, and so the apps a person already has can be read, fixed and reported on without leaving Claude.
Install
Claude and Cowork. Open the Customize menu, browse plugins, choose Lovable. Or install from claude.com/plugins. The first Lovable tool call opens a sign-in window.
Claude Code.
/plugin install lovable@claude-plugins-official
Or from this repository directly:
/plugin marketplace add lovablelabs/mcp
/plugin install lovable@lovable
Restart Claude Code after installing. Verify with /mcp (look for lovable) or ask Claude to list your Lovable workspaces.
Skills
Skills run when your request matches, or directly by name (for example /lovable:launch). Every skill reads the workspace's business profile first, asks before spending credits, reports what each call cost, stops before publishing, and claims only what a tool result shows. The skills were run end to end against the live Lovable MCP before this release; what that changed is in CHANGELOG.md and CONNECTORS.md.
Start here
| Skill | What you say | What you get |
|---|---|---|
know-my-business |
"Tell Lovable about my company" | A short business profile written into workspace knowledge. Every other skill reads it instead of asking again. Spends nothing. |
Get the work done
| Skill | What you say | What you get |
|---|---|---|
launch |
"We're launching Tuesday" | A launch page and changelog on your site, live, plus three posts pointing at it. Copy is written and edited in Claude before anything is built. |
first-customers |
"Find 200 companies like our best customer" | A CRM you own, filled from the tools connected in Claude, every row with its source. |
interview-tracker |
"Talk to ten customers and tell me what they said" | An interview guide, a tracker for every answer, and a summary that quotes people. |
hiring-tracker |
"We're hiring. Track candidates." | A pipeline with stages and a scorecard every interviewer fills the same way. Sign-in required. |
For apps you already have on Lovable
| Skill | What you say | What you get |
|---|---|---|
ask-my-app-data |
"How many signed up this week?" | The answer from the app's live database, with the query shown. Covers weekly numbers and the investor update. Read-only. |
review-my-workspace |
"Check every project for exposed keys and missing RLS" | A ranked table of what needs attention, with the file or table as evidence, for a human to act on. Read-only. |
fix-and-ship |
"Our storage bill tripled. Find out why and fix it." | Root cause with evidence, a plan, the fix applied, and live once you approve. Rollback commit included. |
document-my-app |
"Write the brief for this app" | A brief, spec, README or handover grounded in the code and schema, and the project knowledge kept current. |
New to Lovable
| Skill | What you say | What you get |
|---|---|---|
build-from-this-conversation |
"Make this real" | The idea already worked out in Claude becomes a live app with sign-in, a database and a URL. |
find-what-to-build |
"Look at my clients and what I do every week" | The repeatable work found in your calendar, inbox and sheets, three tools proposed, one built if you pick it. |
spreadsheet-to-app |
"This sheet runs the business. Make it an app." | The sheet as an app, its formulas as rules, its rows in a real database, counts verified. |
claude-design-to-app |
"Turn this Claude Design into the real thing" | The design imported as is, with data, sign-in and actions added behind it. |
Commands
Three commands for people who want the direct route. They predate the skills and stay as they are.
| Command | What it does |
|---|---|
/lovable:build <description> |
Create a project from a prompt, follow the build, optionally deploy |
/lovable:iterate <project> - <change> |
Send a change request, plan first for big ones, review the diff |
/lovable:db <project> [SQL] |
Check or enable a project's database and run SQL with confirmation |
Example workflows
From an idea in Claude to a URL
Make this real
build-from-this-conversation writes the brief from the conversation, you confirm, Lovable builds it in one call, Claude checks each promised feature exists, one follow-up closes gaps, and it publishes when you say so.
Monday numbers
How many people signed up last week, and how does that compare to the week before?
ask-my-app-data reads the schema, writes the SELECT, shows it, runs it for both weeks. Nothing is written and nothing is spent.
Something broke
Our storage bill tripled this month. Find out why and fix it.
fix-and-ship reads the edit history and the database sizes, names the commit and the table, proposes the fix, applies it in plan-then-apply with a yes for each, and deploys on a third yes.
Credits and safety
Four calls spend Lovable credits: create_project, send_message, an approved respond_to_approval, and import-claude-design-from-url. Every skill says what it is about to build, quotes the likely cost (about 2 credits for a static page, 7 to 11 for an app with a database and sign-in), and waits for a yes, one yes per call. Every skill stops before deploy_project, before any database write, and before creating a second project. deploy_project publishes a public URL on Free and Pro plans. Sign-in alone does not limit who gets in, since anyone can create an account by default, so apps that hold personal data are built with an email allowlist naming the domains or addresses that may sign up, enforced in the database policies rather than only in the page, and checked in pg_policies after the build. See CONNECTORS.md for the full table.
Two things are called Lovable
This plugin's MCP server builds and manages projects. A published Lovable app can also expose its own MCP server for the data inside it; that is a separate connector you add in Claude from the app's Agent integrations menu. The skills name which one they mean and never assume an app's own server is connected.
Works in
Claude (web and desktop), Cowork, and Claude Code. Skills that need to upload a file (spreadsheet-to-app) do so directly in Claude Code and Cowork; in Claude chat they pass the sheet's structure and sample rows and ask you to drop the file into the Lovable editor for the rest. The skill says which path it used.
For admins
Enterprise workspaces enable third-party MCP clients under Settings, Security, Privacy and security before anyone can sign in from Claude. The server acts with each person's own Lovable permissions. Every skill already asks before it spends credits, publishes, writes data or changes who can edit a project. To make those stops enforced rather than only promised, set these tools to "ask" in Claude's connector settings: create_project, send_message, respond_to_approval, import-claude-design-from-url, deploy_project, set_project_visibility, set_workspace_knowledge and set_project_knowledge. query_database runs read-only SELECTs as well as writes, so setting it to "ask" also prompts on every read.
What this plugin runs and sends
The plugin connects to one server, the Lovable MCP server at https://mcp.lovable.dev, and talks to nothing else. It runs no local code: the skills and commands are Markdown instructions for Claude, and .mcp.json points at the remote server.
What reaches Lovable is what a skill sends through that server under your own Lovable sign-in: build and edit requests, files you choose to attach, SQL queries against your own project databases, and publish or visibility changes after you say yes. Lovable handles that data under its privacy policy at https://lovable.dev/privacy
scripts/ and evals/ are maintainer tooling. scripts/generate-skills.py writes the skill files and evals/ holds test cases. Neither runs when you install or use the plugin.
MCP server reference
Endpoint
https://mcp.lovable.dev
Transport: Streamable HTTP.
Authentication
OAuth 2.1. MCP clients discover the authorization server via RFC 9728 metadata at https://mcp.lovable.dev/.well-known/oauth-protected-resource and obtain a bearer token automatically. The token is passed as Authorization: Bearer <token>. Connecting grants access to your whole Lovable account, not a single project.
Other clients
Claude, Claude Code and ChatGPT need only the URL. Any other client must also pass Lovable's public OAuth client ID (6d465f583e1e4ce5801b1616f735670c).
Claude Desktop config file:
{
"mcpServers": {
"lovable": {
"type": "http",
"url": "https://mcp.lovable.dev"
}
}
}
Cursor / Windsurf (~/.cursor/mcp.json, or Windsurf's mcp_config.json):
{
"mcpServers": {
"lovable": {
"type": "http",
"url": "https://mcp.lovable.dev",
"auth": {
"CLIENT_ID": "6d465f583e1e4ce5801b1616f735670c"
}
}
}
}
VS Code:
{
"servers": {
"lovable": {
"type": "http",
"url": "https://mcp.lovable.dev",
"auth": {
"CLIENT_ID": "6d465f583e1e4ce5801b1616f735670c"
}
}
}
}
Codex CLI:
codex mcp add lovable --url https://mcp.lovable.dev
codex mcp login lovable
ChatGPT: add the URL directly, no client ID needed.
Tools
- Projects and workspaces: list, create, deploy, remix, inspect; manage visibility
- Agent interaction: send messages to a project's AI agent and retrieve responses
- Code inspection: diffs, file tree, file contents, edit history
- Knowledge: read and write project and workspace AI instructions
- Cloud database: enable a Postgres database, check status, run SQL, get connection info
- Connectors and MCP servers: list, add and remove external integrations
- Templates and libraries: browse template and library projects
- Analytics: historical and real-time project traffic
- File uploads: generate upload URLs for attaching files
create_project and send_message consume Lovable build credits. deploy_project publishes a public URL. query_database runs SQL directly against your project's database.
Full tool reference: Lovable MCP documentation.
Common workflow
1. get_me() → workspace_id
2. create_project(workspace_id, ...) → project_id (uses credits)
3. send_message(project_id, "Add ...") → message_id (uses credits)
4. get_diff(project_id, message_id) → review changes
5. deploy_project(project_id) → live URL
Most tools need a workspace_id; call get_me first. read_file takes a git ref; get latest_commit_sha from get_project.
Troubleshooting
| Symptom | Fix |
|---|---|
401 Unauthorized or expired token |
Re-run your client's login flow to refresh the OAuth token |
| OAuth window loops or never completes | Confirm the client passes the auth.CLIENT_ID shown above |
| "Transport not supported" | The client must support Streamable HTTP |
Tool fails on a missing workspace_id |
Call get_me first and pass the returned workspace ID |
lovable not shown in /mcp |
Confirm the plugin is enabled (/plugin), then restart Claude Code |
Registry
This repository hosts the server.json entry for the MCP registry.
License
Yorumlar (0)
Yorum birakmak icin giris yap.
Yorum birakSonuc bulunamadi