How to Connect Claude to Your Project Schedule With MCP

How do you connect Claude to a project schedule?

Add https://mcp.ganttfather.com as a remote HTTP MCP server, complete the GanttFather OAuth consent, and enable the connector in Claude. Claude can then list projects, read tasks, show a Gantt chart, inspect overdue work, and—when you grant write access—create or update tasks and dependencies.

The current connection is account-wide: it follows every GanttFather project you own or collaborate on, including projects you join later. You do not need to install a local bridge or expose an API key to Claude when using OAuth.

Setup verified: August 7, 2026. Claude and GanttFather can change their connector screens, so use the live GanttFather MCP documentation if a label has moved.

What do you need before starting?

You need a GanttFather account, at least one accessible project, and a Claude client that supports remote MCP over Streamable HTTP. Claude, Claude Desktop, and Claude Code all support remote connections, but their setup screens differ.

Start with read-only access if Claude only needs to summarize a plan. Grant read/write only when Claude must change dates, assignments, tasks, resources, or dependencies.

How do you connect Claude or Claude Desktop with OAuth?

Remote custom connectors are configured through the Claude account, so the same connection can be used by Claude on the web and Claude Desktop.

1. In Claude, open Customize → Connectors.
2. Select Add custom connector. On Team or Enterprise, an Owner must first add the connector under Organization settings → Connectors.
3. Enter https://mcp.ganttfather.com as the remote MCP server URL.
4. Save the connector and select Connect.
5. Sign in to GanttFather and review the requested read-only or read/write permission.
6. In a conversation, use the + → Connectors menu to enable GanttFather.
7. Ask Claude to list the GanttFather projects it can access.

OAuth discovery happens from the server URL. Do not append the old /mcp path: GanttFather’s current Streamable HTTP endpoint is mapped at the domain root.

How do you connect Claude Code with OAuth?

Run this command in a terminal:

claude mcp add --transport http ganttfather https://mcp.ganttfather.com

Then open Claude Code and run:

/mcp

Choose GanttFather and complete the browser sign-in. Anthropic recommends remote HTTP for cloud MCP servers and uses /mcp inside Claude Code to complete OAuth; in this command, /mcp is a Claude Code command, not part of the GanttFather URL.

For a shared .mcp.json or user configuration, use an explicit HTTP type:

{
  "mcpServers": {
    "ganttfather": {
      "type": "streamable-http",
      "url": "https://mcp.ganttfather.com"
    }
  }
}

Claude Code currently accepts streamable-http as an alias for http. Do not omit type in Claude Code JSON; current Claude Code treats a URL-only entry as invalid.

When should you use a bearer token instead?

Use a bearer token when the MCP client cannot complete OAuth, when a headless workflow needs a fixed credential, or when you want a separately revocable connection. In GanttFather, open Settings → AI Agents & API, create a token, select read-only or read/write, and copy the value when it is shown.

For Claude Code:

claude mcp add --transport http ganttfather https://mcp.ganttfather.com \
  --header "Authorization: Bearer YOUR_AGENT_TOKEN"

The equivalent JSON is:

{
  "mcpServers": {
    "ganttfather": {
      "type": "streamable-http",
      "url": "https://mcp.ganttfather.com",
      "headers": {
        "Authorization": "Bearer YOUR_AGENT_TOKEN"
      }
    }
  }
}

Never commit this token to a repository. Prefer an environment-variable substitution when a client supports it, and revoke the token immediately if it appears in logs, screenshots, or source control.

How do you verify that Claude is connected correctly?

Use a read-before-write test. These prompts establish the accessible scope before Claude changes anything:

1. “List my GanttFather projects and do not make changes.”
2. “Open the launch project and summarize its dates, progress, and overdue tasks.”
3. “Show the launch project as a Gantt chart.”
4. “Propose a dependency-safe reschedule, but wait for my approval before updating tasks.”

If the first prompt returns the expected projects, authentication and account scope work. If the third prompt renders a chart or returns a static fallback, visualization works. Only test a write after confirming the correct project by name and ID.

What can Claude do after the connection works?

GanttFather supports FS, SS, FF, and SF dependency relationships with lag through MCP. One current limitation matters for analysis: the MCP critical-path calculation considers finish-to-start edges only, even though the application stores and renders all four relationship types. Treat Claude’s critical-path answer as advisory when the schedule relies on SS, FF, SF, or lag.

Claude also cannot configure, trigger, or inspect the GanttFather Azure DevOps integration through MCP. An Owner or Admin must manage reviewed Pull and Push inside the app.

Which first write prompt is safest?

Ask Claude to inspect, propose, and confirm before mutating:

Read the “Website launch” project. Propose the smallest schedule change that keeps launch on September 30. Show the tasks, old dates, new dates, and dependency impact. Do not update GanttFather until I explicitly approve.

After review, approve a bounded change: “Apply only rows 2 and 3 from that proposal.” This pattern is safer than “fix my schedule” because it names the project, constrains the outcome, and separates analysis from execution.

How do you fix common connection errors?

What security limits should you keep in place?

Use the smallest permission that completes the job. Keep destructive or broad changes behind a human confirmation, name the intended project in every write request, and review Claude’s tool approval prompt before allowing it.

Disconnect the connector in Claude or revoke it under GanttFather Settings → AI Agents & API when it is no longer needed. The agent appears in the GanttFather team surface, and its writes are attributed so collaborators can distinguish automated work from human edits.

When should Claude manage a GanttFather schedule through MCP?

Use this connection when Claude should inspect a live schedule, explain overdue or critical work, visualize the plan, or apply a narrowly approved update without copying tasks into chat. Start with read-only access, name the project in every prompt, and grant read/write only for a specific change. GanttFather’s free tier includes one owned project, two editor seats, and unlimited viewers and guests.

MCP can work with FS, SS, FF, and SF dependencies plus lag, but its current critical-path calculation considers finish-to-start edges only; schedules that rely on the other relationship types need human review. To test the workflow safely, create a free GanttFather project, connect it read-only, and ask Claude to list and summarize the project before approving any write.

What else do people ask about connecting Claude to GanttFather?

Does the connection work with Claude Desktop?

Yes. Anthropic’s remote custom connectors are account-based and available in Claude Desktop as well as Claude on the web. Add the GanttFather URL in Claude’s Connectors settings and authenticate with OAuth. Local claude_desktop_config.json servers are a different mechanism; GanttFather does not require a local process.

Is the GanttFather MCP server free?

Yes. GanttFather includes MCP access on its free tier and does not charge per tool call or per agent. Normal project and editor capacity still applies. The server enforces a limit of 60 requests per minute for each authenticated token.

Can Claude access every GanttFather project?

An account OAuth connection follows the projects the signed-in user can currently access; it does not bypass project membership. Older project-scoped tokens remain limited to their grants. Removing a user from a project removes that account connection’s access to the project.

Can Claude change a schedule without approval?

Claude can invoke write tools when the connection has read/write permission and the client allows the call. For controlled workflows, use read-only access or require approval for each write. Ask Claude to present a dated change table before authorizing a batch of updates.

Does this also work with other MCP clients?

Yes. The same Streamable HTTP endpoint works with compatible clients such as Cursor, Windsurf, VS Code, ChatGPT, and Gemini CLI. Configuration and OAuth support differ by client, so begin with the generic snippets in the MCP documentation and follow the client’s current remote-MCP instructions.

---

Connection behavior checked: August 7, 2026.

Sources

1. GanttFather, MCP server documentation
2. Anthropic, Connect Claude Code to tools via MCP
3. Claude Help Center, Get started with custom connectors using remote MCP
4. Model Context Protocol, Server concepts