Skip to content

Build with AI agents

An AI agent can do anything in your workspace that you can, but only with the access you give it. It acts as you, and the audit log records everything it does.

There are three ways to connect one, from the least setup to the most:

Your agent Use
Claude, ChatGPT or another app with MCP The MCP server. Nothing to install.
A coding agent with a terminal The CLI, with --json.
Your own software The API or the SDK.

Add https://mcp.reminix.com/mcp to your agent as a remote MCP server. The first time it connects, you sign in through your browser and choose one workspace and what the agent may do there. It can’t do anything else. MCP server has the details and the safeguards.

Claude Code and Codex work well with the CLI, and so does any coding agent with a terminal, such as Cursor, GitHub Copilot, Windsurf, Gemini CLI, Antigravity or OpenCode:

Terminal window
npm install -g @reminix/cli
reminix login # you approve it in the browser
reminix workspace list --json
reminix workspace webhook-endpoints list --workspace acme --json

With --json, the agent gets exactly the JSON the API returns. Errors go to stderr as JSON, and the exit code is 0 on success and 1 on failure.

In CI, or in a sandbox with no browser to sign in with, set REMINIX_TOKEN to a personal access token instead. Give the token only the access the agent needs.

The agent skill teaches an agent everything on this page: how to sign in, how to choose the workspace, every operation, the error codes, and what to leave to a person. It also covers the everyday work: finding and running a script, checking a run, and publishing one. We generate the skill from the API, so it never goes out of date.

The quickest way to use it is to tell your agent to read it. This works in any coding agent and needs nothing installed:

Read https://reminix.com/docs/skill/SKILL.md and follow it.

Publish a script with your agent walks through publishing one of your scripts this way.

To have the skill in every session, install it:

Terminal window
npx skills add reminixhq/reminix-skills

The command installs the skill for the agents you pick, such as Claude Code or Codex (every agent it supports). Run it again to update.

With the GitHub CLI:

Terminal window
gh skill install reminixhq/reminix-skills reminix

In Claude Code, you can also install the skill as a plugin. The plugin connects the MCP server too:

/plugin marketplace add reminixhq/reminix-skills
/plugin install reminix@reminix-skills

The skill lives in reminixhq/reminix-skills on GitHub, and we update it with every release. You can also read the files directly: SKILL.md, tools.md and cli.md.

If you don’t write code, you can still work in your workspace through the agent you already use: Claude, ChatGPT, or an agent that keeps working while you’re away.

Use Reminix with your AI agent shows how, step by step.

  1. Add a custom connector with the address https://mcp.reminix.com/mcp, and sign in when the agent asks. In Claude and ChatGPT, connectors are in the settings.
  2. Download the skill and add it to the agent’s skills. In Claude and ChatGPT, skills are in the settings too. On a team plan, an owner can add the skill for everyone.

Then ask for what you need in your own words, such as “Refund $40 on order A-10423.” If the agent works while you’re away, the skill tells it to ask for your approval and carry on instead of waiting.

The docs also come as plain text. /docs/llms.txt lists every page, and /docs/llms-full.txt has all of them in one file. The site’s /llms.txt links to both, and to the API, the MCP server and the skill.

Even with full access, an agent can’t hand out access by itself: to create an API key, it asks, and you approve or deny the request in the app. An agent connected through MCP also asks before it sends your workspace’s data somewhere new, by adding or changing a webhook endpoint. A coding agent that uses your own token or the command line makes webhook changes as you. Approvals explains how. If the action creates a secret, the agent never sees it.

Agents connected through MCP can run scripts in any environment that allows them. They can’t publish a version on their own. They can’t add, change or delete environments, or create schedules or triggers. An agent can ask for a webhook trigger, or a new secret for one, and a workspace owner decides. An agent can rename a script, but it can’t change who may run the script or which of its runs need approval.

An owner or admin can keep agents out of an environment, such as production, or require approval for their runs there. When an environment’s approval setting is Every run needs approval, an agent’s run always waits for a person, even when an owner connected the agent.

The CLI and the API can’t tell a coding agent from you. Give it a personal access token limited to what it needs.