Run one docker compose command across every compose project in a directory tree.

mdc discovers compose.yaml, compose.yml, docker-compose.yaml, and docker-compose.yml, picks one canonical file per directory, and runs the same docker compose command for each target.

# Homebrew
brew install aaronflorey/tap/mdc

# Go
go install github.com/aaronflorey/mdc@latest

Prebuilt binaries are also published on the GitHub releases page.

mdc ps
mdc up -d
mdc --depth 2 pull
mdc --jobs 4 logs --tail 50
mdc --ansi never ps

mdc [mdc flags] <docker compose args...>

• --depth: scan depth (0 = current directory only, default 1)
• --jobs: max concurrent docker compose processes (0 = all targets)
• --quiet-targets: suppress per-target section headers; merged tables, live prefixes, and completion labels retain attribution

git clone https://github.com/aaronflorey/mdc.git
cd mdc
mise install
task setup
task ci

• Discovery starts in the current directory and walks child directories up to --depth.
• If a directory has multiple compose files, this precedence is used:
  1. compose.yaml
  2. compose.yml
  3. docker-compose.yaml
  4. docker-compose.yml
• mdc preserves the target directory basename as the Compose project name when it is unique, and switches to a deterministic unique name only when duplicate basenames would collide.
• Most non-ps commands print grouped output per project in sorted directory order.
• Top-level multi-target detached mdc up -d keeps one live status line per target on TTYs so concurrent Docker progress stays stable; non-TTY output falls back to grouped summaries, and failed targets still print grouped details. Attached multi-target mdc up keeps live target-prefixed output.
• Top-level multi-target mdc logs without follow mode (-f / --follow) uses grouped output for easier historical scanning, while explicit follow mode keeps live interleaved prefixed lines for per-target attribution.
• Top-level multi-target mdc build (parallel mode) suppresses noisy successful BuildKit logs and prints concise [target] build complete summaries; failed targets still print detailed stdout/stderr plus the standard failure summary.
• Top-level mdc events is serial-only for multi-target runs; if multiple targets are discovered, rerun with --jobs 1.
• mdc ps tries docker compose ps --format json, merges results into one table, and falls back to stitched text when JSON is unavailable.
• Top-level multi-target mdc images emits one merged table when all targets share a conservatively parseable table shape; otherwise it falls back to a deterministic stitched view that keeps one header and preserves all rows in sorted target order.
• Exit code is non-zero if any compose target fails.
• Ctrl-C cancels the shared run context so in-flight docker compose processes stop together.
• mdc only parses its own flags (--depth, --jobs, --quiet-targets) and passes everything else through to docker compose.

Development is driven by mise for tool versions and Task for commands. mise install provisions Go, GoReleaser, hk, golangci-lint, and staticcheck pinned to the versions in mise.toml.

mise install        # install pinned tools
task setup          # tidy modules and install git hooks
task test
task test:integration
task lint
task build
task ci

Git hooks are managed by hk. task setup runs hk install --mise so pre-commit lint/format and commit-msg conventional-commit checks are wired against the mise-managed tools.

task run -- ps forwards arguments to go run ..

task test:integration requires a working local Docker Engine with docker compose available.

Releases use release-please for versioning and changelog management. Merged Conventional Commits are collected into a release PR, and merging that PR creates a vX.Y.Z tag, publishes release archives with GoReleaser, and updates the aaronflorey/homebrew-tap formula.

• Contributing Guide
• Security Policy
• Code of Conduct
• MIT License