Cursor
To put toolkit files in Cursor, use Set up and open. Starting Cursor Agent is a separate operation and does not write those files.
Before you start
Section titled “Before you start”Install Cursor on this computer, connect Code Config, and choose a toolkit and local project folder. Review the toolkit’s Cursor compatibility before writing into an existing project.
Open your project in Cursor
Section titled “Open your project in Cursor”In the Code Config app, select your project, toolkit, and Cursor among the desktop apps. Choose Set up and open.
Code Config writes the supported files into your project and asks your computer to open that folder in your own Cursor. It does not choose a Cursor sign-in or create a managed Cursor IDE session.
If Cursor does not open, check the result. Your toolkit files are in the project. Open that folder in Cursor to use them. means writing succeeded but opening the app did not. Open the folder yourself instead of repeating the write unnecessarily.
Check your tools
Section titled “Check your tools”In Cursor, confirm the correct folder is open, then find and invoke a selected skill or command. Check the reported name; conversion may have renamed it. Read Notes on this setup for skipped fields or tools.
What Code Config changes
Section titled “What Code Config changes”| Content | Project destination |
|---|---|
| Skills | .agents/skills/ |
| Agents | .cursor/agents/ |
| Commands | Converted to skills marked for user invocation. |
| MCP servers | .cursor/mcp.json |
A record under .code-config/ identifies the files Code Config owns. Subsequent setup uses that record to distinguish its files from yours.
Each toolkit MCP server the project gets is one entry in .cursor/mcp.json, under the server’s own name. A server that runs on this computer has its command and arguments; one reached over the internet has its address. A key either kind needs is written right into that entry, under env or headers. Because a project is usually a git repository, Code Config also adds .cursor/mcp.json to that repository’s own private ignore list (not the .gitignore you would commit) the first time it writes there, so the file and any key inside it are never committed. Entries you put there yourself are left exactly as they are, and an entry Code Config wrote for a server your toolkit no longer carries is taken back out. Restart Cursor to pick up the change; it does not notice a rewritten file while it is already running.
Names may be normalized to lowercase letters, numbers, and single hyphens. Tool allowlists, command argument hints, and command shell syntax can lose behavior during conversion; the result names those limitations.
Cursor Agent
Section titled “Cursor Agent”From a terminal in the project:
csc launch cursorThis starts the managed Cursor Agent terminal run with the account you choose, and writes the toolkit’s MCP servers into the project’s .cursor/mcp.json the same way Set up and open does, except that these entries come back out again once the run ends. It does not deliver toolkit skills, agents, commands, or hooks. Use the desktop Set up and open operation first to put those in the project. There is no equivalent Cursor project-setup command under csc launch.
Limitations
Section titled “Limitations”A server whose key is not set up on this computer yet, or whose program this computer does not have, is left out of .cursor/mcp.json and named in the result with the command that fixes it. If .cursor/mcp.json is already tracked by git, no server that needs a key is written into it at all: each one is left out and named the same way, though a server that needs no key is still written.
Code Config’s Cursor setup does not write rules, context files, output styles, hooks, plugins, repository settings files, workflows, or marketplace entries.
cursor_project_destination_unavailable means no local project folder was available. cursor_project_config_sync_failed means writing did not complete. These are write failures, unlike a successful write followed by an app-opening problem.