Hi PH 👋
Here, one of the creators behind the collab. We built it so our team could stop bouncing between Slack, GitHub, Notion and half a dozen separate agent tools. An agent, sitting on each channel that the team already resides on, is backed by any MCP server.
Here’s how we use it ourselves. Collab joins our Slack and Telegram bots, with Notion MCP and GitHub MCP behind them. Within our workgroup, anyone (devs or not) can ping the bot to view code, review a feature, or file an issue. In the community group, users ask @Kollab to report bugs or ask how something works, and each message goes directly to our backend board through the Notion MCP. Expressions used to get lost in DMs; Now it is not.
What we have underestimated the most is scheduled work. We thought we were sending a digest job, but a scheduled job on Collab is really a time agent. The same cron can call any MCP tool, pull from the knowledge base, run as a specific agent role, and post back to any channel. Ours right now: someone generates a weekly changelog from GitHub issues, someone cross-checks our status page against Sentry, someone pings the on-call before standup. Same thing under the hood, totally different jobs on top.
When we need more than a quick response, AgentCore is there. Long running with its own file system and a browser. And since skills are just regular GitHub repos, anything the team iterates on turns into a skill that the whole org can install by name. We’re still beginners on this part, and this is probably where we’ll find the weirdest uses.
Question for PH: If you had an agent sitting on your team’s channels with full MCP access, what would be the first defined task or skill you would write? I don’t know what people will think. So far the answers have been all over the map, and two of them are already in our next release.