The Modyl connector

Modyl watches the agents you run, such as Claude Code and Codex, against the rules you set for them. The connector lets you ask ChatGPT or Claude about that work in plain words: which sessions ran, what they changed, what a standing rule refused, and which escalations are waiting for you. You can answer an escalation from the same chat.

It is a remote MCP server at https://apex.modyl.to/mcp. You sign in with an account through WorkOS. The first time you sign in, Modyl makes a space of your own, and only you can read it.

Connect it in Claude

1. Open Settings, then Connectors, and choose Add custom connector.
2. Name it Modyl and enter https://apex.modyl.to/mcp. Leave the request headers empty.
3. Choose Add, then Connect, and sign in when Claude sends you to the Modyl sign-in page.
4. In a conversation, open the + menu, then Connectors, and switch Modyl on.

Connect it in ChatGPT

1. Go to chatgpt.com/plugins, select the plus button, then Add custom MCP server. Custom servers need developer mode, which your plan or workspace policy must allow.
2. Name it Modyl and, under Connection, enter https://apex.modyl.to/mcp. Choose OAuth for authentication.
3. Confirm the warning, select Create as a plugin, and sign in when ChatGPT sends you to the Modyl sign-in page.

Your first session

Your space is empty until an agent sends something to it. Until then, every answer says what to connect and how. Three things can fill it:

• This chat. Every call ChatGPT or Claude makes to the connector is recorded in your space from the first call on, with the tool, its arguments and which app made it.
• Claude Code. The Modyl plugin for Claude Code records each prompt, tool call and file your agent touches.
• Codex. Modyl hooks into Codex the same way.

Claude Code and Codex send their activity with an intake key for your account. Write to hello@modyl.to from the address you signed in with, and we will send you a key and the install steps.

What to ask

• What are my agents doing right now?
• What did my agents do in the last four hours?
• Was anything refused, and why?
• Is any escalation waiting for me? Settle it as a false alarm.
• Show me everything the release session touched.

What it can do

Almost every tool only reads. Each one reads your own space and nothing else.

• Read activity: agentsNow, activitySince, sessionTrace, traceSearch and activityExport answer what your agents did, from the newest session down to every record of one session.
• Read refusals and escalations: refusals lists the events where a standing rule fired, and escalationsWaiting lists what is waiting for a person.
• Answer an escalation: settleEscalation records your verdict, worth raising or a false alarm, with an optional note. It is the tool that changes what your agents may do next, so keep it asking for approval every time.
• Look at another session: peek, peekAgents and routeAsk search what other agents in your space did. The btw tools pass a short message to another agent session in your space.

In ChatGPT and Claude, which show MCP Apps, waiting escalations and a session timeline show inline, with buttons to answer an escalation.

Escalation events

Modyl announces each new escalation as the MCP event apex.escalation.raised. A ChatGPT Work chat, on the web or in the desktop app with Cloud selected, can subscribe to it, and Modyl delivers each escalation to that chat as it is raised. Claude does not take events from a connector yet, so in Claude you ask whether anything is waiting.

Your data

Modyl keeps every call the connector receives, including the tool name, every argument and the name of the app that called it, as activity in your own space. The privacy policy says what is kept, who processes it and for how long, and the terms cover the connector.

To disconnect, remove Modyl from your connectors in ChatGPT or Claude. To have your space deleted, write to hello@modyl.to from the address you signed in with.