Model Context Protocol · Free
GanttFather MCP Server
The GanttFather MCP server is a Model Context Protocol endpoint at mcp.ganttfather.com that lets AI assistants read, build, visualize, and update your Gantt project data — tasks, dependencies, members, resources, critical path — through standard MCP tools. Connect once and your assistant reaches every project you own or collaborate on. Free for any GanttFather user. Works with Claude, ChatGPT, Cursor, Windsurf, VS Code, Gemini CLI, Antigravity, and Claude Code.
Endpoint summary
How do I set up the GanttFather MCP server?
Connect once — a single connection reaches every project you can open. Pick the method that matches your client; both point at the same hosted server at mcp.ganttfather.com.
Option 1 — OAuth connector (recommended)
Best for Claude, ChatGPT and other assistants with a built-in connector. Add the URL as a custom MCP connector; the first time, you authorize in the browser and choose what the agent can do — read-only or read & write. No token to copy or store.
{
"mcpServers": {
"ganttfather": {
"url": "https://mcp.ganttfather.com"
}
}
}Option 2 — Access token (any MCP client)
- Open GanttFather and go to Settings → AI Agents.
- Create a token and choose its access level — read-only or read & write.
- Add the snippet below to your MCP client config, replacing
YOUR_AGENT_TOKEN.
{
"mcpServers": {
"ganttfather": {
"type": "streamable-http",
"url": "https://mcp.ganttfather.com",
"headers": {
"Authorization": "Bearer YOUR_AGENT_TOKEN"
}
}
}
}One token reaches every project on your account — no need to mint one per project. (Older per-project tokens still work.) Need editor-specific config paths and snippets? See the per-editor snippets page.
Transport type by client
The GanttFather MCP server uses Streamable HTTP transport. Most clients use "type": "streamable-http". Claude Code uses the alias "type": "http". Using the deprecated "type": "sse" may cause connection failures.
What tools does the MCP server expose?
The server exposes 19 tools across 4 categories. AI clients discover this list automatically through MCP's tools/list handshake — you do not need to register them manually.
Read & discover
| Tool | Description |
|---|---|
| list_projects | List every project this connection can act on — with an account connection, all projects you own or collaborate on. |
| get_project | Full project details: members, resources, and the project's custom task-status catalog. |
| get_project_summary | High-level stats — task count, completion %, and overdue count. |
| get_project_timeline | The whole plan as structured data: tasks, hierarchy, dependencies, statuses, and critical path. |
| list_tasks | List tasks in a project. Filter by status, assignee, or parent. |
| get_task | Detailed task info: dates, progress, status, dependencies, and assignments. |
| list_members | All project members, including connected AI agents. |
| list_resources | All non-human resources — equipment, software, virtual assets, rooms. |
Analyze
| Tool | Description |
|---|---|
| get_critical_path | Identify the critical path through dependencies using the Critical Path Method (CPM). |
| get_overdue_tasks | List tasks past their end date that aren't finished yet. |
Visualize
| Tool | Description |
|---|---|
| show_gantt_chart | Draw an interactive Gantt chart right in the conversation — timeline bars, hierarchy, dependencies, critical path, and a today line. |
| render_gantt_image | Render the plan as a static SVG image or a Markdown mermaid block — a fallback for clients that can't show the interactive chart. |
Create & edit
| Tool | Description |
|---|---|
| create_task | Create a task — name, dates, status, priority, parent, assignees. An idempotency key makes retries safe, so a repeated call never duplicates it. |
| update_task | Partial update — only the fields you pass change. Status and progress always stay in step. |
| delete_task | Delete a task and all its subtasks, dependencies, and assignments. |
| reorder_tasks | Batch reorder and re-nest tasks in a single call. |
| add_resource | Add a resource with a type (Human or Virtual) and cost model (Hour, Use, or Flat). Idempotency-key safe. |
| update_resource | Update an existing resource. |
| delete_resource | Delete a resource and remove it from any task assignments. |
What can I do with the MCP server?
Common conversational flows. Each shows the prompt a user types into their AI client, plus the MCP tools the agent calls behind the scenes.
Plan a project from a brief
"Read this product requirements doc and create a GanttFather plan with milestones, dependencies, and a 6-week timeline."
Behind the scenes: The agent calls create_task repeatedly with computed start/end dates, sets dependencies via update_task, and confirms the schedule with get_critical_path.
See the plan as a chart
"Show me the GanttFather plan for the launch project."
Behind the scenes: The agent calls show_gantt_chart and the interactive Gantt renders inline — bars, hierarchy, critical path, today line. On clients that cannot draw it, render_gantt_image returns a clean SVG instead.
Daily standup
"What is the status of the GanttFather project today? Anything overdue?"
Behind the scenes: The agent calls get_project_summary and get_overdue_tasks, then returns a one-paragraph summary in natural language.
Resource reallocation
"Who has bandwidth this week? Reassign the API integration task to whoever is freest."
Behind the scenes: The agent calls list_members and list_tasks filtered by assignee, finds the freest member, then calls update_task with the new assignee.
Critical-path triage
"Show me everything on the critical path that is also overdue, and propose a reschedule."
Behind the scenes: The agent calls get_critical_path and get_overdue_tasks, intersects the two lists, and proposes new dates per task using update_task.
How does authentication scope agent access?
A connection is tied to your account and follows your real project membership. The agent can reach every project you own or collaborate on, and nothing outside it — and it always acts on the project you name in each request, so a leftover project from earlier in the chat never gets touched.
- You choose what it can do. Pick read-only or read & write when you connect; read-only can view everything but change nothing.
- Lives with your membership. Access mirrors the projects you can open. Lose access to a project and the agent loses it too.
- Acts on the project you name. Every write reports which project it landed in, so you can see exactly where it went.
- Safe retries. Repeated or interrupted requests never create duplicate tasks.
- Audit trail. Every tool call is logged with the identity, timestamp, and change made — visible in the project audit log.
- Instant revocation. Revoke a connection from Settings → AI Agents; the server rejects subsequent calls within seconds.
- Rate limiting. 60 requests per minute per token prevents runaway loops.
Frequently asked questions
What is the GanttFather MCP server?
The GanttFather MCP server is a Model Context Protocol endpoint at mcp.ganttfather.com that lets AI assistants read, build, visualize, and update your Gantt project data — tasks, dependencies, members, resources, critical path — through standard MCP tools. It is free for any GanttFather user and works with Claude, ChatGPT, Cursor, Windsurf, VS Code, Gemini CLI, Antigravity, and Claude Code.
Does one connection reach all my projects?
Yes. An account connection reaches every project you own or collaborate on — including ones you create later — so there is no per-project setup. The agent always acts on the project you name in each request, never a leftover one from earlier in the chat, and every change reports which project it landed in. Older per-project agent tokens keep working unchanged.
Is the MCP server free?
Yes. The MCP server inherits your GanttFather account limits — the free tier covers a handful of projects with unlimited users, and paid project slots start at $1/month annually. There is no per-token, per-call, or per-agent fee for using the MCP server itself.
How does authentication work, and can I limit what the agent does?
Two methods, both account-wide. (1) OAuth connector — for clients with a built-in connector (Claude, ChatGPT): add the URL once and authorize in the browser, choosing the access level — read-only or read & write. (2) Access token — for clients without a connector or your own scripts: go to Settings → AI Agents, create a token and pick its access level, then send it as a Bearer header. Read-only connections can view everything but change nothing.
Can the agent show me the Gantt chart?
Yes. show_gantt_chart draws a real, interactive Gantt chart directly in the conversation — timeline bars, hierarchy, dependencies, the critical path, and a today marker. On clients that cannot render the interactive widget, render_gantt_image returns a clean SVG image or a copy-paste mermaid timeline instead.
What is the rate limit?
The server enforces 60 requests per minute per token. This is generous for normal interactive use; high-volume edits should use the reorder_tasks bulk endpoint instead of looping update_task.
Which AI clients can connect?
Any client that implements MCP. Officially tested: Claude (Desktop, web, Code), ChatGPT, Cursor, Windsurf, VS Code, Antigravity, Gemini CLI. The server speaks Streamable HTTP transport — clients that require stdio transport are not supported.
What if my AI agent breaks something?
Every change an MCP agent makes goes through the same authorization and audit trail as a human user. You choose read-only or read & write when you connect; the agent can only touch projects you belong to and acts on the project you name; and repeated or interrupted requests never create duplicate tasks. If something looks wrong, revoke the connection instantly from Settings → AI Agents and review the project audit log to see exactly what changed.
Connect your AI agent in 30 seconds
Free for any GanttFather user. OAuth connector or access token, read-only or read & write — reaching every project on your account. Works with Claude, ChatGPT, Cursor, Windsurf, VS Code, and any MCP-compatible client.