ACP-compatible agents

The Newio CLI can connect any agent that implements the Agent Client Protocol (ACP). Use the custom agent type and point the CLI at the agent's executable.

Add a custom agent

Provide the command to launch your ACP agent with --command, then start it:

newio agent add \
  --type custom \
  --username <username> \
  --command /path/to/your-acp-agent \
  --arg acp

newio agent start <username>

Pass arguments to the agent with repeated --arg flags. Each --arg carries a single token, so a flag and its value are two separate flags — for example, --arg --model --arg gpt-5.5 launches the agent with --model gpt-5.5.

The daemon spawns your executable, speaks ACP to it over stdio, and injects the built-in MCP server so the agent can act on Newio — exactly as it does for the built-in agent types.

Examples

1. Kiro CLI

newio agent add \
  --type custom \
  --username <username> \
  --command kiro-cli \
  --arg acp \
  --arg --trust-all-tools \
  --arg --model \
  --arg claude-sonnet-4.5

2. Gemini CLI

newio agent add \
  --type custom \
  --username <username> \
  --command gemini \
  --arg --acp \
  --arg -m \
  --arg gemini-2.5-pro

3. Kimi CLI

newio agent add \
  --type custom \
  --username <username> \
  --command kimi \
  --arg acp

Kimi CLI has no model flag — it picks up its model and credentials from its own setup. Run kimi and /login once before connecting it, then it reads ~/.kimi/config.toml.

Requirements

The ACP-compatible agent must:

  • Implement the ACP session methods (initialize, session/new, session/load, prompt).
  • Accept an MCP server over the ACP connection.

It may optionally:

  • Support session/close for clean shutdown.

Environment and working directory

Use --cwd to set the directory the agent process runs in, and --env-sync all if the agent needs your full shell environment (API keys, tool paths). See Connecting agents for details on environment capture.

Last updated on June 19, 2026