# A schedule an AI can edit

**Is there a scheduling tool an AI can edit by itself?** Yes: PalmaPlan exposes every command of its scheduling engine to an AI agent, over MCP. Claude or another assistant adds rows, moves blocks, links tasks, sets milestones — with your rights, signed in its name, undoable —, and everyone who has the schedule open sees the change live.

## In short

- **The schedule itself**, not a description: the AI reads and writes the real schedule, the one the team uses.
- **With your rights**, never more: owner, editor or reader, the role is checked on every call.
- **Signed and undoable**: every write carries the agent's name in the History, and `Ctrl+Z` stays yours.
- **Try before applying**: a simulation answers "what if…?" without a single date moving.
- **Any MCP client**: the protocol is open; Claude Code and Claude are the ones that have been tried.

## What an AI can do in it

- **Build** a whole schedule from a description: phases, rows, blocks, links, in one call.
- **Edit it**: add, rename, move, lengthen, indent, link, set a milestone, colour, comment.
- **Analyse it**: critical path, delays, risks from absences, what changed between two versions.
- **Track it**: a block's actions, their owners — a person or an AI —, the derived progress.
- **Share it and get it out**: invite, publish, export in the v23 format, as CSV, an image or a PDF.

The [tools reference](/docs/en/agents/tools-reference/) gives the full list, generated from the server itself.

## The safeguards

- **Your consent, first.** An agent connects through OAuth: a consent page names the client and what it asks for. It leaves with a **named token** in its own name, which you revoke in one click; revoking also closes its current session.
- **Read or write.** A read-only token can read, export, simulate; if it tries to write, it gets an explicit refusal.
- **Your rights, no more.** The agent acts under your role on each schedule; a schedule that was not shared with you, it does not see.
- **Nothing on your account.** No tool touches the password, the profile, the export or the deletion of the account: those gestures stay the person's.
- **Everything is traced.** The History says who did what, when — "Claude (you)" for what your agent wrote.

## And the team meanwhile

The schedule stays a Gantt chart you drive with the mouse, alone or together. When the agent writes, every open screen follows over a realtime channel, without reloading; if two writes cross, the conflict is reported instead of one silently overwriting the other. An AI becomes one team member among others, whose every move is visible.

## Going further

- [Tokens and scopes](/docs/en/agents/tokens-and-scopes/) — read, write, revocation, what an agent cannot do.
- [Making a Gantt chart with Claude](/docs/en/faq/gantt-chart-with-claude/) — the step by step.
- [Project management and MCP](/docs/en/faq/project-management-and-mcp/) — why an open protocol.
- [History and Compare](/docs/en/guides/history-and-compare/) — reading back what changed.
- [For agents](/docs/en/agents/for-agents/) — the machine path, for the agent reading this page itself.