Docs

Use kits from your agent

Your agent can pick the kit, show you the plan and create the project for you. Ask in your own words: "create a landing page with animations and analytics".

It runs on your machine with npx -y create-first-commit mcp and it's the only one that creates projects. Before running anything, the agent shows you the plan (plan_init) and needs your approval.

Claude Code:

$ claude mcp add first-commit -- npx -y create-first-commit mcp

Cursor (.cursor/mcp.json):

{
  "mcpServers": {
    "first-commit": {
      "command": "npx",
      "args": [
        "-y",
        "create-first-commit",
        "mcp"
      ]
    }
  }
}

Codex (~/.codex/config.toml):

[mcp_servers.first-commit]
command = "npx"
args = ["-y", "create-first-commit", "mcp"]

Creating a project takes a few minutes. If your agent cuts off long-running tools, raise its timeout (in Claude Code, the MCP_TOOL_TIMEOUT variable, in milliseconds).

search_kits
Searches kits by text, category or library (stripe, sentry, motion…).
get_kit
A kit's options with their values and requirements, skills and default command.
recommend_kit
Kit and options for your project's description, with the reasons. Always a valid combination.
plan_init
Everything init would do (commands, files, skills), without running anything.
init_kit
Creates the project. Local server only, with explicit approval.
add_library
Adds a catalog library or a skill to the project. Local server only.

Nothing to install: https://first-commit.dev/mcp (HTTP, no session). It searches, recommends and plans; it doesn't run commands, so it returns the command for your agent to run in your terminal.

$ claude mcp add --transport http first-commit https://first-commit.dev/mcp

The first-commit skill teaches the agent when to use kits, how to choose options and how to maintain the project with update, add and doctor. Every official kit installs it.

$ npx skills@latest add Marlon-Chaviano/first-commit --skill first-commit

The skill's source: skills/first-commit.

On every kit's page, the Prompt tab next to the command gives you a prompt to paste into your agent with the options you chose: the agent runs the command with --dry-run, shows you the plan and applies it once you confirm.