Create your first team
You don't fill in a form — you describe the workflow in plain words and an agent composes it. Then you refine by talking.
1. Describe it
mmt new
It opens your editor. Write how your team works, in one paragraph. Example:
a strategist plans the work, a coder implements it and opens a PR and updates
the ticket, a reviewer comments on github and loops with the coder until
approved, then qa runs the tests
Save and close. The agent composes the team and shows it:
spec-to-prod take an agreed spec to prod — plan, build, review loop, qa, audit (no spec challenging; we already know what we want)
1. strategist
2. coder ·github-pr ·ticket-status
┌ loop · until reviewer approves · max 3
│ 3. reviewer ·github-comment
│ 4. coder ·github-pr
└
5. qa ·run-tests
Everything you described — the members, the skills, the review loop — is inferred. You do not add a "lead"; it's implicit.
2. Refine by talking
At the prompt, describe any change (or press enter to save):
add a security reviewer after the reviewer, only comments on github
The team redraws. Keep going until it's right, then save.
You can also give the description inline:
mmt new "researcher gathers sources, writer drafts, editor loops with the writer until clean"
Use --local to save it to the current project's ./teams/ instead of your global library.
3. Look and run
mmt show my-team
mmt run my-team "a real task" --sim # free preview
mmt run my-team "a real task" # for real, on your subscription
Edit an existing team
Same idea, by talking:
mmt edit spec-to-prod "run the coder on opus and the reviewer on sonnet"
mmt edit spec-to-prod "make qa a loop that retries until tests pass"
Pick models per member
Add a model: to any member, or override a whole run:
mmt run spec-to-prod "…" --model sonnet
Precedence: --model (run) → member model: → team model: → your default. See Running.
Next
- Give members real capabilities → Skills
- Repeat-until-good → Loops & gates
- Share it → Sharing