AI-assisted integration
Let your coding agent integrate OhAuthCat: ohauthcat agent install adds the MCP server and integration rules, and ohauthcat init adapts to your existing project.
You can let a coding agent (Claude Code, Cursor, Windsurf or GitHub Copilot in VS Code) add OhAuthCat to your app. The CLI gives the agent the context it needs and makes the code changes reviewable and reversible; it does not run a model itself.
1. Install the agent files
Section titled “1. Install the agent files”Run ohauthcat agent install in the project. Without --editor, it sets up the editors whose folders exist in the project. For each editor it writes:
| Editor | MCP server entry | Rules or skill file |
|---|---|---|
claude-code |
.mcp.json |
. |
cursor |
.cursor/mcp.json |
. |
windsurf |
~/ (global) |
. |
vscode |
.vscode/mcp.json |
. |
The MCP entry points at the hosted MCP server and sends your key from the OHAUTHCAT_API_KEY environment variable; set that variable in the environment your editor starts from. The rules file teaches the integration steps and the security rules: verify state, key users on the profile’s idp_id, never trust the email alone, and handle dsync.user.deleted.
The change is shown as a diff and written after you confirm, or with --yes. --dry-run only shows it. Running the command again changes nothing when the files are current. The editor configuration formats follow each editor’s documentation; check that your editor version picks them up.
2. Let the agent run init
Section titled “2. Let the agent run init”Ask the agent to add SSO with OhAuthCat. Following the rules, it runs ohauthcat init --dry-run, shows you the plan and applies it with your approval. init adapts to the existing project, keeps the auth you already have, records every change and can be reversed with ohauthcat init --undo. See the init quickstart.
What the CLI reads
Section titled “What the CLI reads”Neither command reads .env files or sends your code anywhere. See what init and agent install read.
Compared with other providers
Section titled “Compared with other providers”As observed in their public products and docs in October 2026: WorkOS offers a CLI command that edits your code with AI, and the Scalekit setup command installs coding-agent plugins and an MCP server. OhAuthCat’s approach differs in two ways: the CLI calls no model, so the changes are deterministic and the same on every run, and every change init makes is recorded and can be undone with one command. The coding agent you already use does the reasoning, with OhAuthCat’s MCP server and rules as its context.