looped.shDocs
Looped MCP

Connecting an agent

Connect Claude Code, Codex, Claude.ai, ChatGPT, Cursor and other MCP clients to Looped, and install the Looped skill

Every client connects to the same server:

https://api.looped.sh/mcp

It's a Streamable HTTP server that signs in with OAuth. The first time your client connects, it opens a Looped sign-in in your browser. You sign in the way you normally do, approve the client, and you're back in your agent. You don't need an API key.

The MCP page in Manage (under your company settings) has each client's steps ready to copy. It also has a Copy setup prompt button: paste that prompt into any agent and it sets itself up, then checks the connection by listing your companies.

Claude Code

Add the server:

bash
claude mcp add --transport http looped https://api.looped.sh/mcp

Then run /mcp in Claude Code, pick looped and sign in. Install the skill as well (see below).

Codex

Add the server, then sign in when Codex asks:

bash
codex mcp add looped --url https://api.looped.sh/mcp

Claude.ai and the Claude apps

Open Settings › Connectors, add a custom connector, paste the server URL and sign in to Looped. The connector is then available in Claude on the web, desktop and mobile. Claude Duet uses the same connectors.

ChatGPT

Open Plugins, add a custom MCP server, paste the server URL and sign in to Looped. On a Business or Enterprise workspace, an admin might need to add it for you. OpenAI Dots use the same connection.

Cursor, VS Code and other clients

Add Looped to the client's MCP configuration:

json
{
  "mcpServers": {
    "looped": { "type": "http", "url": "https://api.looped.sh/mcp" }
  }
}

For agents with their own connector screen, such as Grok, add a custom connector with the server URL and sign in, or send them the setup prompt from the MCP page.

The Looped skill

The skill is a page of instructions the agent reads before it works in Looped. It covers how Looped's amounts and dates work, which actions need your go-ahead, what the agent must never ask you for, and the playbooks for longer jobs. Agents make fewer mistakes with it, so install it alongside the server wherever your client supports skills.

For Claude Code:

bash
mkdir -p ~/.claude/skills/looped && curl -fsSL https://app.looped.sh/skills/looped/SKILL.md -o ~/.claude/skills/looped/SKILL.md

For Codex:

bash
mkdir -p ~/.codex/skills/looped && curl -fsSL https://app.looped.sh/skills/looped/SKILL.md -o ~/.codex/skills/looped/SKILL.md

The skill changes as Looped does, so run the command again now and then to pick up the latest. Clients without skills get the essentials from the server itself when they connect.

Checking it works

Ask the agent "Which Looped companies can you see?" It should list every company you belong to. If it can't find any Looped tools, the client isn't connected yet: in Claude Code run /mcp, and in other clients check the connector is switched on for the chat you're in.

Are you an AI? Visit llms.txt — these docs as plain markdown.

On this page