ManicTime already tracks what you do on your computer automatically (which apps you use, which documents you open, which websites you visit, when you’re active/away).

In the new version, ManicTime adds a new option: an AI connection using something called an MCP server. This lets an AI tool understand your recorded activity and turn it into organized, timesheet-ready time entries without you having to manually sort and label everything.

What does this mean in practice?

Until now, ManicTime:

• automatically recorded activity
• but you usually had to manually assign that time to projects/tasks (for example "Project A”, "Support”, "Meeting”"

With the AI/MCP integration:

• the AI can review your recorded activity (window titles, documents, URLs, app usage)
• it can infer what you were working on (for example "Support work”, "Client project B”, "Internal meeting”"
• it can propose structured time blocks and apply tags so your timesheet is mostly done for you

Result: less manual work, faster timesheets, and more consistent tagging.

How does the AI connection work?

ManicTime does not lock you into a single built-in AI provider. Instead, it uses the MCP standard, which means you can connect ManicTime to many AI tools that support MCP (for example Claude, ChatGPT desktop-style tools, GitHub Copilot in VS Code, Codex, and others).

Once connected, you can ask questions like:

• "When did I start working yesterday, and how long was I active?"
• "How much time did I spend on support work, including related coding and browser activity?" And the AI can respond with a clear breakdown and, if you allow it, write tags back into ManicTime.

What happens to my data?

The MCP server runs locally inside the ManicTime client on your computer. ManicTime does not send your activity data through ManicTime servers for this feature.

However, the data you choose to analyze is sent from your computer to the AI tool/provider you connect. Whether that AI processes data locally or in the cloud depends on the tool and its settings.

Why is AI useful?

• Saves time: less manual sorting and tagging
• Improves accuracy: the AI can use context to categorize work more consistently
• Fits into your workflow: can be combined with tools like Jira and used for reporting or invoicing

How to set it up

ManicTime's MCP server is built into the client. There's nothing extra to install - your AI tool simply runs the ManicTime executable with the mcp argument.

ManicTime client default executable paths

If you installed ManicTime elsewhere, adjust the path accordingly.

Connect from popular AI tools

Claude Code (CLI)

claude mcp add manictime-client --transport stdio -- \
  "/Applications/ManicTime.app/Contents/MacOS/ManicTime" mcp

Claude Desktop

Open Settings → Developer → Edit Config, then add:

{
  "mcpServers": {
    "manictime-client": {
      "command": "/Applications/ManicTime.app/Contents/MacOS/ManicTime",
      "args": ["mcp"]
    }
  }
}

Restart Claude Desktop.

OpenAI Codex

codex mcp add manictime-client -- \
  /Applications/ManicTime.app/Contents/MacOS/ManicTime mcp

Visual Studio Code (GitHub Copilot)

Option A (UI):

1. Ctrl+Shift+P / Cmd+Shift+P → MCP: Add Server

2. Select Command (stdio)

3. Enter the executable path followed by mcp, for example:

   Windows:

    "C:\Program Files\ManicTime\ManicTime.exe" mcp
   

   Mac:

    "/Applications/ManicTime.app/Contents/MacOS/ManicTime" mcp
   

4. For the name, enter ManicTime

Option B (config file):

Create .vscode/mcp.json:

{
  "servers": {
    "ManicTime": {
      "type": "stdio",
      "command": "C:\\Program Files\\ManicTime\\ManicTime.exe",
      "args": ["mcp"]
    }
  }
}

Then open Copilot Chat and switch to Agent mode to query tools.

In January 2026 MCP support in some AI tools is still a bit buggy. If the connection doesn’t work right away, try the following:

• If you’re using Claude or ChatGPT, explicitly remind it that you’re using the desktop app (with MCP enabled), not the web version.

• Ask it to connect to the MCP server again (sometimes it takes a couple of tries before it starts working).

• If it still doesn’t connect, restart the AI app and ManicTime and try again.

What you can do with it

Once connected, your AI can access ManicTime activity for any time range:

• Computer usage
• Applications
• Documents and websites
• Tags

In addition to reading, it can also generate tags.

We are sure you will have much better ideas, but here is an exchange with Codex:

Connect to ManicTime mcp and list available services

• Connected successfully.

Configured MCP server:

• manictime-client (/Applications/ManicTime.app/Contents/MacOS/ManicTime mcp)
• Status: enabled

Available ManicTime MCP services in this session:

1. get_timeline_and_summary_types
2. get_timelines
3. get_environments
4. get_groups
5. get_group_summary
6. get_combined_activities
7. get_combined_activity_summary
8. get_total_duration
9. save_tag_activities

I also confirmed live connectivity by calling get_timeline_and_summary_types successfully.

When did I come to work yesterday and how much time was I active?

For yesterday (Wednesday, February 25, 2026):

• First active at 07:59:47
• Total active time: 6h 58m

We use GrooveHQ for customer support, but support work also involves coding tools, browser, and other apps. Can you figure out how much time I actually spent on support yesterday, including those related activities? Also show me the start and end times for each support session.

Here's what I found for February 25, 2026:

• Direct GrooveHQ time: 1h 01m
• Total inferred support time: 2h 52m

I inferred support sessions by merging GrooveHQ activity with surrounding tool usage, bridging gaps of up to 20 minutes - so coding or browser work between ticket actions is included.

#	Start	End	Duration
1	08:03	08:06	3m
2	08:41	09:02	21m
3	09:34	09:35	1m
4	10:16	10:16	<1m
5	10:43	10:43	<1m
6	10:54	11:13	19m
7	11:54	12:53	59m
8	13:24	13:24	<1m
9	13:39	13:46	7m
10	14:13	14:23	10m
11	15:33	16:18	44m
12	16:45	16:52	7m

Create those as tags named "ManicTime, Support"

Done. I added all 12 intervals as ManicTime, Support tags using DoubleTag mode, so your existing tags are untouched.

Verified total for 2026-02-25: 2h 52m ✓

The key thing to notice here: ManicTime captured everything automatically in the background. The AI read the raw activity, reasoned about context, and wrote structured tags back - no manual entry required.

How can you try it?

1. Update to ManicTime v2026.1 (or newer).
2. Choose your AI tool (Claude, ChatGPT Desktop, Copilot/VS Code, Codex...).
3. Configure MCP using the executable path table above.
4. Start with read-only questions, then move to AI-assisted tagging and write-back.

From raw activity to a tagged timesheet - that's automatic time tracking, powered by whichever AI you choose.

You can get ManicTime v2026.1 Beta here:

• Windows
• Mac
• Linux

Community MCP server implementation

Alongside the built-in MCP server in ManicTime v2026.1, a community member also created an independent MCP server implementation:

• https://github.com/yllibed/manictime-mcp

This is a great example of how open MCP tooling enables alternative integrations and custom workflows around ManicTime data.

If you want to explore or contribute to a community-driven approach, check the repository for setup details and current capabilities.

Share your ManicTime MCP workflows

MCP opens up possibilities we haven't even thought of yet. Beyond the example above, you could use it to reconcile time against Jira tickets, generate client-ready summaries, spot patterns in how your day actually unfolds versus how you planned it or build automated end-of-day reports.

We'd love to hear how you're putting it to work and especially what you're missing.