@one-agent/act

Deterministic MCP tool runner for ONE.

Within this repository's Re in Act implementation, @one-agent/act is the external action layer for the optional act() interface: it is responsible for deterministic tool execution while higher-level agent logic decides when to call act() inside the RAS.

Relevant public reference:

• Working Draft Specification

Important boundary: the Re in Act spec requires reason() and makes act() optional. one-act is this repository's MCP-backed implementation of that optional action interface.

one-act does not have a model auth flow. one auth configures the main one agent model, and reason auth configures the model used by reason(). one-act only reads MCP/tool server configuration.

MCP Config

one-act reads MCP server config from ~/.config/one/act.json or ONE_ACT_MCP_SERVERS.

{
  "mcpServers": {
    "chrome-devtools": {
      "transportType": "stdio",
      "command": "npx",
      "args": ["-y", "chrome-devtools-mcp@latest", "--autoConnect"],
      "env": {
        "FOO": "bar"
      },
      "daemon": true
    },
    "remote-http": {
      "transportType": "streamable-http",
      "url": "https://example.com/mcp",
      "headers": {
        "Authorization": "Bearer token"
      }
    }
  }
}

Supported transport shapes follow @mcpc-tech/core McpServerConfig. one-act only recognizes per-server daemon: true: marked servers stay resident in the daemon, and unmarked servers connect on demand for each call. The daemon strips the flag before passing runtime config into mcpc.

Reference:

• https://github.com/mcpc-tech/mcpc/blob/main/packages/core/src/service/tools.ts