Install preset in your AI coding tool
Connect Claude Code, Cursor, or Codex to the preset gateway so your agent reads your system before it generates.
The preset gateway exposes your system to AI tools as a set of tools they can call: read the tokens, check a snippet against the rules, resolve a value. Any client that speaks the Model Context Protocol can connect. The steps are the same for Claude Code, Cursor, Codex, and any MCP-aware client.
Connect over OAuth
OAuth is the recommended way in. The client opens a browser, you approve access once, and the token refreshes itself. No key is pasted into a config file. Add the gateway as a remote MCP server:
{
"mcpServers": {
"preset": { "url": "https://mcp.presetai.dev/mcp", "transport": "http" }
}
}On the first call the client opens a browser to approve access. Approve it, and the connection is live.
Verify it
Ask the agent to read the system. If a check references your own forbidden rules and scope, the agent is reading the real system. If it returns generic defaults, the session fell back to unauthenticated: re-run the OAuth step.
Headless and CI
CI has no browser, so use a project key set as an environment variable instead. Keep it in your secret store, never in the repo. The same tools are available either way; only the auth changes. Per-tool setup and the full tool list are in Integrations and the docs.
Connected? Now watch it work: see Catch drift from AI-generated UI.