Work locally — init_local_project and the slim MCP
Move from a conversation to work on your machine with the CLI, and when to use the slim webmati mcp instead of the full MCP.
init_local_project: from the MCP to the CLI
Working over the MCP and want your own editor, git, or Claude Code / Cursor / Codex
on your machine? Tell the AI: "I want to work on this website locally". It calls
the init_local_project tool and gives you:
- the command to install the CLI on macOS/Linux and Windows;
- the sign-in command
webmati login --web …(you approve the key in your browser — the MCP never creates or sees that key); - the command to clone the website into a folder;
- a zip link of that very project, valid for one hour;
- the work loop: validate → test → push → preview → publish.
Carry on in the CLI — start with Install.
The slim webmati mcp or the full MCP?
With the CLI you also get an MCP that runs on your machine: webmati mcp. It
has just 8 tools:
search_tools describe_tool call_tool execute status pull push validate
The full MCP (https://mcp.webmati.com/mcp) | The slim MCP (webmati mcp) | |
|---|---|---|
| Runs | on Webmati's server | on your machine, in the project folder |
| Tools | 182, all loaded into the context | 8; finds a tool when needed (search_tools, describe_tool) |
| Size of the tool list | 100% | 97.4% smaller |
| Works on | the website, through tools | the website and the local project (status, pull, push, validate) |
| Suits | conversations, small jobs, nothing to install | coding agents, long jobs, saving tokens |
The tool list is what an agent reads on every turn, so the slim MCP saves tokens from the first turn. Measure it yourself:
webmati mcp --measure
How to add webmati mcp to Claude Code and Cursor: see
AI agents.