Built-in tools: from files to the browser

Tools are provided by the LoneAct service running on your computer, and the model only sees capabilities that are registered and allowed. Every tool call stays in the conversation as a card with its input, status, and result.

Files

Read, search, and edit files inside the project; every path is relative to the current project folder.

Terminal

Run commands, install dependencies, and run tests inside the project boundary, with output returned to the conversation.

Browser

Open pages, read content, click, and fill forms; you can authorize it explicitly or attach it to the current browser.

Remote and device tools

Call external capabilities exposed by authorised devices, MCP servers, or connectors.

Execution follows approval

When writing files, running terminal commands, controlling the browser, or calling sensitive external services, LoneAct asks for your confirmation according to tool policy. Approval is a condition of execution, not a prompt you can bypass.

Read the target and parameters

Before approving, check the tool's target directory, command, browser site, and parameters. Approve only the scope the current task really needs; when unsure, deny first and ask the Agent for a smaller step.

MCP and connectors

MCP servers turn external systems into tools the Agent can call; connectors provide richer external service context. Both are managed in Settings → Integrations, and the tool list and status update live.

MCP servers

Add, enable, refresh, and authorize MCP servers; only tools that are exposed successfully enter a request.

Connectors

Connect external services, check authorization status, and use them on demand in a conversation.

Tools catalog

See the tools actually registered in the current workspace, not a static feature list.

Device tools

See what the current computer offers to remote callers.

Skills: turn domain knowledge into reusable capability

Skills are reusable packages of domain knowledge that can include working methods, checklists, templates, and specialized steps. Loading them for an Agent Team is more reliable than repeating the same explanation in every prompt.

Good for capturing team conventions, code review checklists, release processes, and business knowledge.
A role can load only the skills it needs, reducing irrelevant context.
A skill is not a tool; it changes how the Agent works, while actual execution is still bounded by tools and permissions.

Start with one small skill

Write one stable team rule as a skill, prove it on a real task, and expand only afterwards instead of maintaining a large knowledge base from day one.

Plugins and Hooks

Plugins

Package skills, rules, Hooks, and MCP configuration into an installable capability set that can be enabled or disabled in a workspace.

Hooks

Run controlled commands before or after conversations, tool calls, and other events for checks, prompts, or automation.

Plugins and Hooks extend workflows and also widen the execution surface. Keep the version and source of team-shared plugins traceable, and check which commands and external services they reach before enabling them.