AgentsRoom tools settings

  • Desktop app
  • All plans
View as Markdown

What it does

Every agent AgentsRoom launches gets the AgentsRoom MCP tools: backlog, project memory, Agent Mail, agent consoles, teams, SSH, databases, the QA Bot and more, about 155 tools. Settings > AgentsRoom tools lets you switch off the families (or single tools) you do not use, for example because you run your own memory, backlog or delegation system. A switched-off tool is not even listed to the agent: its schema costs no context, the instructions about it leave the agent's startup prompt, and a call to it is refused with a message that says it was switched off on purpose. You can also rewrite the description of any tool (what the model reads to decide when and how to call it).

Where to find it

  • Settings > AgentsRoom tools: the switches for every project (account level, synced across your computers), and the description editor.
  • Project settings > Capabilities > AgentsRoom tools > Customize: this project's overrides.
  • Folder settings > Agents > AgentsRoom tools > Customize: the overrides of every project filed in the folder.
  • Edit agent > advanced options > AgentsRoom tools > Customize: one agent's overrides.
  • The plug button of an agent's header > all tools > Choose the tools: shortcut to the settings page.

How to use it

  1. Open Settings > AgentsRoom tools. The top bar shows how much the tools an agent receives weigh (measured on the real tools), out of the total.
  2. Switch a family off (Backlog, Project Memory, Agent Mail, Agent consoles, Teams, Remote machines, Databases, Idea Radar, Triggers, QA Bot...). Unfold a family to switch single tools off (for example backlog_delete).
  3. Click a tool name to read its full description and its parameters. On the Settings page, "Edit the description" replaces the text the model reads for all your agents; "Back to the AgentsRoom text" removes your version.
  4. Start or restart the agents: a change applies at their next start.

In a project, folder or agent editor, each switch shows "Inherited" until you set it there; the round-arrow button hands it back to the level above. The narrowest level wins, switch by switch: agent, then project, then folder (and its parent folders), then Settings. A tool switch never brings back a tool whose family is off at the same or a wider level.

Settings

  • mcpToolToggles (global, Settings > AgentsRoom tools): object of familyIdOrToolName: true | false, absent key = on. Synced with the account.
  • mcpToolToggles (folder override, Folder settings > Agents > AgentsRoom tools).
  • mcpToolToggles (project override, Project settings > Capabilities > AgentsRoom tools).
  • mcpToolToggles (agent override, Edit agent > advanced options > AgentsRoom tools).
  • mcpToolDescriptions (global only, Settings > AgentsRoom tools > a tool > Edit the description): your own descriptions, synced with the account.
  • Family ids: backlog, memory, notes, todos, prompts, skills, marketplace, commands, mail, delegation, agents, teams, projects, ssh, databases, kubernetes, secrets, ideas, triggers, inboxes, settings, usage, product, desktop, qa.
  • Project settings > Capabilities > Project Memory off also switches the memory tools off for that project's agents.

Agent tools (MCP)

  • settings_get / settings_set with key: "mcpToolToggles" at the global, folder and project scopes, and mcpToolDescriptions at the global scope.
  • agents_save (mcpToolToggles) and settings_set with scope: "agent" for one saved agent.
  • capabilities_get lists only the tools the agent really has.

Providers

All agent CLIs with an MCP client (Claude Code, Codex, Copilot, OpenCode, Antigravity, Grok, Cursor, Kimi, Mistral Vibe, Amp, oh-my-pi and the others): the AgentsRoom MCP server itself leaves the switched-off tools out, whatever the CLI. Claude Code may load tool schemas on demand, so the context gain is largest on CLIs that load every schema at each turn (Codex, OpenCode...).

Mobile

Not on the phone: the settings are edited on the desktop. An agent started from the phone still gets the switches of its project and of its agent.

Limits

  • Applies to agents started or restarted after the change.
  • Always on: capabilities_get, session_update, ask_user (governed by Structured questions) and git_set_commit_message (governed by Settings > Git).
  • The embedded browser and Chrome tools are granted per agent (Capabilities), not here.
  • An agent running on a remote host over SSH keeps every tool.
  • Agent Mail switched off for one agent only: another agent can still write to it, and it may be told that mail arrived, but it cannot read it. Switch Agent Mail off for every agent to avoid this.
  • A rewritten description replaces the description only; parameters keep the AgentsRoom text. When an update changes the original text, the tool shows "Updated by AgentsRoom" and lets you compare, keep yours or go back.

Common questions

  • Can I turn off the AgentsRoom memory or backlog tools because I use my own? Yes: Settings > AgentsRoom tools, switch the family off. The agents no longer see those tools nor the instructions about them.
  • Can I give one agent fewer tools than the others? Yes: Edit agent > advanced options > AgentsRoom tools > Customize.
  • Can I stop agents from deleting tickets? Yes: unfold Backlog and switch backlog_delete off.
  • Can I change what an AgentsRoom tool description says? Yes, on the Settings page: click the tool, Edit the description.
  • Are the AgentsRoom tools split into several MCP servers I can disable? They stay one server (fewer processes), but each family can be switched off, with the same effect for the agent.
  • How do I see which tools an agent actually has? The plug button of its header lists every tool it loaded.

Related