MakerKit MCP 2.0: Your AI Assistant Now Manages Your Entire Project

This post is part of a unique release, including the MakerKit Plugin System and MakerKit CLI v2.

MakerKit's MCP server launched last year with three tool categories and twelve tools. That was a starting point but MCP 2.0 is a much more powerful tool.

66 tools. Every kit variant. Plus the CLI is now a separate MCP server with 10 more tools.

Your AI assistant no longer guesses at your database schema, invents components that already exist, or asks you to run commands it doesn't know about. It starts your dev services, inspects your database, manages your translations, reads your emails, installs plugins, and resolves merge conflicts — all through MCP.

The MCP Server is designed to be used with any MCP-compatible AI Agent - such as Claude Code, Cursor, Codex or Gemini. While choosing the right tool to use is much up to the agent itself, the MCP Server does its best to provide the best experience possible for AI Agents working with MakerKit.

This is also part of a larger effort to make Makerkit more vibe-codeable. More exciting developments coming soon!

Kit MCP Server: 56 tools

Database schema inspection (6 tools)

Your AI assistant can inspect your database without reading raw SQL or guessing at table structures.

Instead of reading migration files (which show historical changes, not current state), the AI reads your schema directly. It knows what columns exist, what foreign keys connect tables, and what enum values are valid — right now.

Before: "Add a status column to the projects table." The AI reads 15 migration files, tries to reconstruct the current schema, gets confused by a renamed column in migration 8, and produces incorrect SQL.

After: The AI calls get_table_info({ table: 'projects' }), gets the exact current schema, and writes the correct migration in one shot.

Database operations (4 tools)

Beyond inspection, the AI can operate on your database directly.

After writing a new migration, the AI checks if it applies cleanly, seeds test data, and verifies the result — without you running a single command.

Migration management (4 tools)

The generate_migration tool compares your schema files against existing migrations and produces exactly the diff needed. No manual migration writing.

Environment management (5 tools)

One of the most requested additions. Your AI assistant can now read, validate, and update environment variables.

When the AI installs a plugin that requires API keys, it knows what variables are needed, what format they should be in, and where to put them.

Dev services management (3 tools)

Start, stop, and check your local development stack without touching the terminal.

Tell the AI "start the project" and it calls kit_dev_start with the right services for your variant. No more remembering which combination of pnpm commands to run.

Email inspection (6 tools)

Two tool sets: one for inspecting received emails in your local Mailpit inbox, one for managing email templates.

Mailbox tools (runtime emails):

Template tools (source templates):

Testing email flows just became conversational: "Send a test invite, then show me what the email looks like." The AI triggers the action, reads the received email from Mailpit, and reports back.

Translation management (7 tools)

Full i18n lifecycle — from adding locales to tracking coverage.

"Add Spanish translations for the billing namespace" is now a single instruction. The AI adds the locale, copies the structure from your primary locale, and fills in translations.

UI components (4 tools)

These shipped in v1, but they prevent one of the most common AI coding mistakes: reinventing existing components.

Scripts and code quality (4 tools)

The run_checks tool is the upgrade from get_healthcheck_scripts. Instead of returning a list of commands for the AI to run manually, it executes them directly and returns structured pass/fail results.

Project status and prerequisites (2 tools)

PRD management (10 tools)

Track requirements with structured Product Requirements Documents.

Dependency management (1 tool)

No more running npm outdated and guessing what's safe to upgrade. The advisor categorizes updates by risk level and recommends an upgrade order.

AI prompts (2 prompt templates)

MCP 2.0 also includes structured prompt templates for complex tasks:

• Code Review: Comprehensive review with configurable focus area (security, performance, maintainability, TypeScript, React) and severity level
• Schema Design: Guided schema creation with access patterns and data sensitivity inputs, producing complete schemas with relations and security policies

CLI MCP Server: 10 tools

The kit MCP server gives your AI assistant read access to your project. The CLI MCP server gives it write access: install plugins, create projects, pull updates, resolve conflicts.

Together, the two servers cover the full loop. Your AI assistant can inspect your database schema, decide you need a new migration, create it, run it, install a monitoring plugin to track the new feature, and verify everything passes checks — all without you leaving your editor.

To add the CLI MCP server, drop this into your editor's MCP config:

{
  "mcpServers": {
    "makerkit-cli": {
      "command": "npx",
      "args": ["-y", "@makerkit/cli@latest", "makerkit-cli-mcp"]
    }
  }
}

For the full deep dive on CLI MCP capabilities — including three-way merge conflict resolution — see Introducing the MakerKit CLI v2.

Every kit, one MCP server

MCP 2.0 ships with all four MakerKit variants:

• Next.js + Supabase
• Next.js + Drizzle
• Next.js + Prisma
• React Router + Supabase

The first release only supported Drizzle and Prisma. Now every MakerKit customer gets the full MCP experience.

Setup hasn't changed. Build the server once:

pnpm --filter "@kit/mcp-server" build

Locate the JS file generated and copy the relative path (see .mcp.json as an example). If the MCP Server fails, please try using the absolute path. Different agents may behave differently.

Claude Code

Claude will automatically detect the MCP Server and use it as we ship the configuration in the .mcp.json file.

Codex

Open the Codex YAML config and add the following:

[mcp_servers.makerkit]
command = "node"
args = ["<path-to-mcp-server>"]

Please change the <path-to-mcp-server> to the actual path of the MCP Server.

Cursor

Open the mcp.json config in Cursor and add the following config:

{
  "mcpServers": {
    "makerkit": {
      "command": "node",
      "args": ["<path-to-mcp-server>"]
    }
  }
}

Please change the <path-to-mcp-server> to the actual path of the MCP Server.

Other Agents

Please refer to the documentation for other AI Agents. The format is normally similar with minimal differences.

Why this matters

Most starter kits give you code. MakerKit gives you code and 66 tools for AI to work with that code effectively. As AI-assisted development becomes the default way people build software, the kits that work well with AI assistants will outperform the ones that don't.

What 66 MCP tools mean in practice:

• Fewer hallucinations: The AI reads your actual schema, actual env vars, actual components — not a stale mental model
• Faster onboarding: New team members' AI assistants understand the project immediately
• Safer changes: The AI runs checks automatically, creates proper migrations, and validates environment config
• Less context switching: Start services, install plugins, manage translations, inspect emails — all from your editor
• Full project lifecycle: From makerkit_create_project to run_checks, every stage is AI-accessible

What's next

The kit MCP server and CLI MCP server are two interfaces into the same project. Today you interact with them through your terminal or your AI editor. But MCP tools aren't limited to the command line — any MCP-compatible client can call them. We're building toward that.

For the technical deep dive on how plugins are distributed and wired up, see Introducing the MakerKit Plugin System. For the latest additions to the plugin catalog, see New Plugins: Honeybadger, Directus, and Meshes Analytics.