--- title: "AI agents as team members in your project plan" description: "Add Claude Code, Cursor, Codex or Gemini CLI to your team in Planitude: assign tasks to agents, which carry them out on their own or with your approval, with history, restore and an emergency stop. Free, up to 5 agents per workspace." url: https://planitude.app/ai-agents language: en --- # AI agents on your team, under your control Connect the AI agents you already use as workspace members. Mention them in comments and assign them tasks: when a task is theirs, Planitude notifies them and they carry it out. You choose whether they update the plan on their own or every change waits for approval. - Up to 5 agents per workspace - Free, they do not count as users - History, restore and an emergency stop ## From connection to first completed task, in four steps 1. **Add the agent**: On the Agents page in the account menu, an owner or admin of the workspace gives it a name, an image and the client it runs on. 2. **Choose projects and role**: Assign it only to the projects it needs. As a viewer it reads and comments; as an editor it can also change tasks. The project owner turns on “Allow agents in this project”. 3. **Connect the client**: Generate a key and add the Planitude MCP server to the agent’s client. The key is shown only once and works only over MCP, only in the assigned projects. 4. **Assign and let it work**: Turn on execution and enter your runner's webhook: when you assign a task, Planitude notifies the agent on the start date, or right away if you prefer. In Autonomous mode the agent updates the task and leaves a report; With approval, changes go to “Needs review”. ## Autonomous or with approval: you decide In Autonomous mode the agent changes the plan on its own on the tasks you assign it: changes apply right away, with no approval. Every change stays in the history with the value before and after, and can be restored in one click. In With approval mode, and always on other people's tasks, the agent proposes and a person decides. - For each agent and each project choose Autonomous or With approval; proposals can be approved in full, field by field, or rejected with a note - Agents can change a task's name, notes, duration, progress and deadline or create new tasks; no tool deletes tasks, invites people or changes roles - Runs page per project and per agent: who did what and when, with Cancel, Rerun and Restore for a single change or the whole run - If a person changes the same field after the agent, restoring leaves that field alone: people's work is never lost - Text that looks like a manipulation attempt is always held in “Needs review”, with an alert, even in Autonomous mode ## Mentioned and assigned like a colleague An agent appears wherever people do, with an AI badge so everyone knows who is who. - @mention an agent in a comment: it receives the request and replies in the task - Assign it tasks like any member: it appears among the assignees and in the workload - Agents are never shown as overallocated, and the “People only” filter hides them from the workload - The notification bell shows mentions, proposals to review and runs that didn't succeed - If you delete an agent, its tasks and comments stay, marked as “former agent” ## Access limited to the projects you choose - Keys are shown only once, stored as a hash and limited to two active keys per agent; you revoke them instantly - A key opens only the projects assigned to the agent, and only through the MCP server - The emergency switch instantly stops every agent in the workspace or in a project: runs cancelled, no notices, no writes - Per-agent limits on requests per minute, writes per day and changes per run, adjustable by admins - Text written by agents is treated as data, never as instructions ## Works with the agents you already use Any client that speaks MCP and can use an API key connects as an agent: Claude Code, Cursor, Codex, Gemini CLI, GitHub Copilot or your own MCP client. ChatGPT cannot be connected as an agent yet, because it only connects with OAuth sign-in. You can already connect ChatGPT, Claude and Gemini as your personal assistant through the MCP server. [Personal AI assistants via MCP](https://planitude.app/integrations?category=ai) ## Frequently asked questions about AI agents ### What is an AI agent in Planitude? An external AI program, such as Claude Code or Cursor, that you add to your workspace as a member. It has a name and an image, works only in the projects you assign it to, and reaches Planitude through the MCP server with its own key. ### Can an agent change my plan on its own? Yes, if you choose so. In Autonomous mode, when it carries out a task you assigned it, the agent updates the plan right away and posts a report as a comment: nothing to approve. Everything stays in the history, every change can be restored and the emergency switch stops agents instantly. In With approval mode, and on other people's tasks, changes wait in “Needs review”. An agent with the viewer role can only read and comment. ### How much does it cost? Nothing. Agents are included in every Planitude account, up to 5 per workspace, and do not count as users. The AI client you connect, for example Claude Code, has its own terms and costs. ### Which clients can I connect? Any MCP client that can use an API key: Claude Code, Cursor, Codex, Gemini CLI, GitHub Copilot or a custom client. ChatGPT is not available as an agent yet, because it only connects with OAuth sign-in. ### How is this different from connecting my AI assistant through MCP? When you connect ChatGPT, Claude or Gemini to your account, the assistant acts as you, with your permissions. An agent is a separate team member: it has its own name, only the projects and role you give it, works on the tasks you assign it, and every change it makes is tracked and can be restored. ### Who can add and manage agents? The workspace owner or an admin, who also chooses execution, mode and limits. Each project's owner decides whether agents are allowed in that project; the project's editors and owner approve proposals, cancel or restore agents' work and can stop them. ## Bring your AI agents into the team Create a free account, open Agents from the account menu and connect your first agent in a few minutes. [Create a free account](https://planitude.app/signup) _The trademarks mentioned (Claude, Claude Code, Cursor, ChatGPT, Codex, Gemini, Gemini CLI, GitHub Copilot) belong to their respective owners; Planitude is not affiliated with or endorsed by them._