# Project management and MCP

**What does MCP bring to project management?** MCP (Model Context Protocol) is an open protocol that lets an AI assistant discover a service's tools and use them. In project management, it changes what you hand to the AI: no longer a description of the schedule, but the schedule itself — which it reads, builds, simulates and tracks with your rights. PalmaPlan is built for this: the project management tool connected to AIs, agnostic and full MCP App.

## In short

- **A protocol, not an integration**: any client that speaks MCP over HTTP connects, with no code written for it.
- **The MCP server is the API**: every command of the scheduling engine is a tool, named in the schedule's own words.
- **Full MCP App**: in a client that displays them, such as Claude, the Gantt chart opens and is edited in the conversation.
- **Agnostic**: the schedule depends on no assistant; you change agents without changing tools.
- **Run together**: people and AIs work on the same schedule, each with their own rights, every move signed.

## Why an open protocol

A proprietary integration ties a tool to one assistant. An open protocol does the opposite: PalmaPlan describes its tools once, and every MCP client — Claude Code, Claude, a script, another agent — discovers them on its own (`tools/list`) and calls them with the token you granted it. Authorisation follows OAuth 2.1, with dynamic client registration: nothing to create by hand on your side.

## What it changes for a complex project

A complex project means dozens of rows, dependencies, absences, versions, a team. MCP lets an AI work in it like a team member:

- **when building**: a whole schedule is born from a description, phases and links included;
- **when steering**: "what is late? what is the critical path? what changed since Monday?";
- **when deciding**: "what if acceptance testing slipped two weeks?" — simulated, then applied or not;
- **when tracking**: a block's actions assigned to people or to AIs, the progress derived from what is done.

The Gantt chart stays a Gantt chart a person drives with the mouse, with no AI at all: MCP is one more door, not the engine.

## What PalmaPlan does, concretely

- **Connect**: `claude mcp add --transport http palmaplan https://palmaplan.com/mcp`, or a custom connector in Claude.
- **See**: the Gantt chart, the kanban and the agenda display in a conversation (an MCP App) — the Gantt chart has been tried in Claude, the kanban and the agenda not yet.
- **Stay in control**: one named token per client, revocable, read or write; the History that signs every write; `Ctrl+Z`.
- **Where the data is**: hosted in France, in the European Union. Access is by invitation during the beta ([request one](/en/request-an-invitation/)).

## Going further

- [The agent guide](/docs/en/agents/) — connecting a client, tokens, the embedded view, the tools reference.
- [Making a Gantt chart with Claude](/docs/en/faq/gantt-chart-with-claude/) — the step by step.
- [A schedule an AI can edit](/docs/en/faq/schedule-an-ai-can-edit/) — what it can do in it, and the safeguards.
- [The website's MCP page](/en/mcp/) — the command, and what an agent can do.
- [What we know about you](/en/data/) — the data, where it lives, for how long.