Config

a2a-serve config set takes the same flags as a2a-serve start opencode, plus settings that only config set takes: container limits, network keys, backups, task retention and Google sign-in. The config reference lists them all.

Each change is validated, saved and printed as a diff:

$ a2a-serve config set --sandbox local --allow-tools read,search --secret ANTHROPIC_API_KEY
  sandbox      (default) → local
  secret       + ANTHROPIC_API_KEY (stored)
  allow-tools  + read, + search
✓ saved .a2a/config.yaml
! coder is running; restart to apply sandbox, secret: a2a-serve restart

A running agent applies --description, --private-access, --public-access, --allow-email, --allow-tools, --deny-tools, --disable-mcp and --owner-chat at once, and messengers too. Other settings apply after a2a-serve restart, or pass --restart to config set.

Command	What it does
a2a-serve config show	Shows the settings. --as-command prints the commands that recreate the agent, and -o yaml prints the settings as YAML.
a2a-serve config set <flags>	Changes settings. --dry-run only prints the diff.
a2a-serve config unset <setting> [<entry>…]	Removes entries from a list, or resets a setting to its default.
a2a-serve config edit	Opens config.yaml in your editor and validates it on save.

List settings are keyed: --secret and --env by name, --mount by destination, and other lists by value. Setting an existing entry replaces it. Secrets covers --secret and the flags that take tokens.

config set --agent opencode creates an agent without starting it. Use it when the first start needs settings that start doesn't take, such as --tailscale-authkey.