The issue was basically this: I might be working on Georgi’s account, but his stuff is spread across Gmail, Drive, Notion, Slack, etc. So I kept doing this with agents: “check Gmail for Georgi” “also check Drive” “look in Notion” “see if anything happened in Slack” The agent has access to all of these tools, but I’m still acting as the router every single time. Peon stores the client connection instead. I create Georgi once, attach the apps that belong to him, optionally set boundaries, then I can ask: what’s latest from Georgi? The agent uses Peon to figure out which connected sources belong to him and search across those. One of my tests pulled together an email about a BoxNow integration, files from Drive, project state from Notion, and a related Slack thread into one answer. The part I care about isn’t the aggregation. It’s that I didn’t have to tell the agent where any of it lived. This gets more useful when a client has 10–20 sources. Contacts, folders, channels, repos, whatever. I’m not trying to rebuild Gmail or Slack MCPs. Peon sits on top of Composio for the actual app connections. It’s a routing/context layer in front of them. I’ve been comparing the same tasks with and without the client routing. In my current tests it’s about 18% fewer tokens. I want a much larger set before I lean on that number. It’s free to start: create a client, attach a couple of apps, connect it to Claude / Cursor / ChatGPT. Curious if anyone else working with multiple clients has run into the same problem, or if you’ve solved it differently.