Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Everyday workflows

Run mcpstack import to choose one or both harnesses, then enter a stack file path or HTTP(S) URL. Selecting none reports that nothing was imported and exits without asking for a source. Installed harnesses appear first and are marked (installed); each group is alphabetical. list, show, export, diff, and use select one harness. Pass --harness codex or --harness claude to skip the prompt; repeat --harness only with import to select both. For unattended runs, provide the source or export file and --non-interactive. When importing to both harnesses, each is processed in turn; an error can leave the first harness updated. Use the namespaced form for harness-specific options.

Try it

Download demo.yaml, then preview against an unused config path:

mcpstack validate demo.yaml
mcpstack codex use demo.yaml --config scratch.toml --dry-run --non-interactive

Nothing is written or executed. For Claude Code, use claude and scratch.json.

Share

mcpstack codex export work.yaml --non-interactive
mcpstack claude diff work.yaml
mcpstack claude import work.yaml

import merges servers; those absent from the stack stay configured.

Switch

mcpstack codex use personal.yaml --dry-run
mcpstack codex use personal.yaml

use removes servers absent from the stack. An empty stack clears them all. Unrelated settings are preserved. Repeating an identical import or switch writes nothing.

Both harnesses support these commands. See secrets and permissions and recovery before sharing or applying changes.