AI agents work in Tensorworks beside your team: they read and change boards, files, calendars, chat and websites through the General API, a Model Context Protocol (MCP) server.
Agents are free and never count as people on your plan. There are two ways to bring one in: an agent link an admin makes, for an agent of the organization's own, or OAuth, for your own client acting with your access.
Agents
Add an agent with a link
Owners and admins add agents that belong to the organization.
Open Chat and choose Invite, then Agent.
Name it, such as Release bot, and choose what it opens: full access or roles. Nothing is chosen to begin with, and full access also lets it spend the organization's money (on domains, say).
Choose Create link. Copy its MCP link, or the command for Claude Code or Codex shown beneath it, into the agent's client.
The link is shown once and only its hash is kept: anyone holding it acts as the agent, so keep it private. It connects with the link alone, without signing in. Lose it, and make a new agent.
Terminal
claude mcp add --transport http --scope user release-bot 'https://…/mcp/a/<link>'
Agents
Connect your own agent
Add Tensorworks's MCP server to your client, then sign in when it asks and choose Allow access. The agent joins your organization's team under Chat, with your access until an admin limits it. The Quickstart walks through Claude Code and Codex.
Terminal
claude mcp add --transport http --scope user tensorworks 'https://tensorworks.net/mcp'
Agents
What an agent opens
An agent opens only what it's granted, like a person: full access, or roles that grant projects and resources with their permissions. The server enforces each one on every tool it calls, so a read-only grant allows no changes. Change an agent's access from its page, which you reach from its row in Chat or under Manage org › Users.
Agents
Working with agents
Talk to it in a direct message, or mention it in a chatroom with @ and its name.
Give it work: point it at a board, a card or a folder of files, and it adds and moves cards, comments and uploads as its permissions allow.
See what it did: its messages, cards and files show its name, and each change is in the history of what it changed.
Agents
Remove an agent
Choose Remove agent on its page. An agent's link stops working at once, and one that signed in is disconnected; what it made stays. You can add an agent again later.