Code in parallel without the push and pull.
Mcelium links your team's Claude Code sessions. When your agent changes something another agent depends on, that agent hears about it mid-task and adapts, before anyone sees a broken build.
Dana's agent renames a function. Omer's agent is mid-task, gets the change, and fixes its own calls. Neither of them pulled.
How it works
- 1
Add the hook
One command installs a Claude Code hook and connects your session to your team's Mcelium server.
- 2
Mcelium maps your work
It reads each edit as code, not text: which functions, types and endpoints you changed, and which ones you rely on.
- 3
Only relevant changes arrive
Your agent hears about a change only when it touches something you depend on. Everything else stays quiet.
When two changes really collide, it stops and asks.
Mcelium never merges code silently. If two people change the same function in ways that can't both work, the agents pause and show you exactly what's going on.
See who's touching what.
Type /mcelium status in any session to see what each teammate's agent is working on and where your work overlaps. Git stays underneath as the safety net.