Hello, this is 3Rabbitz.

Last time, we introduced how AI tools find, proofread, and add content to ManualWorks documents through MCP. This time, it's tasks.

Look back on your day, and you'll see this a lot. You open a task, copy its content, paste it into AI, and when the work is done, you summarize the result, paste it back as an answer, and change the category. Some days, copying things over takes more effort than the actual work.

What if AI read the task itself, left an answer when the work was done, and closed the task for you? Connect the ManualWorks MCP, and you can work that way in Claude Code, Codex, and Gemini CLI. Most of the process of reading tasks and recording results can happen through a conversation with AI.

Here are eight scenes that show what actually changes.

Start with one sentence: “Handle this task”

Case 1. From a single task ID to implementation

A bug task has been assigned to you. It has several answers with the customer's inquiry and a colleague's analysis.

Conversation with AI

User: Read the following task, including its answers, and fix it. 7a81162c9432d394

AI: This is the task "The web viewer looks empty when the index has a page range entry."

I read the 2 answers as well. The cause is the part that converts range entries into web locations.

I'll check the related code and fix it.

...

I've finished the fix and the tests. Here are the changes.

AI reads the customer's inquiry and the colleague's analysis all at once. You'll fix the wrong place less often because of context lost while copying and pasting.

Case 2. AI summarizes, people check

When the work is done, you need to record what you changed and why. The conversation you worked in already has all the evidence, so leave the summary to AI.

User: Leave what we've done so far as an answer on this task.

AI: I've posted the following answer.

What was done

- Cause: ...

- Fix: ...

- Tests: 39 related tests passed

Answers are written in standard Markdown, so headings, lists, tables, and code show up as they are. If you ask AI to leave a record just for yourself as a memo, other people won't see it.

Case 3. Read customer emails and draft answers

With a mail channel, emails from customers come in as tasks. Several inquiry emails have piled up again today.

User: Read the tasks in the "Customer Emails" filter, find supporting content in the user guide, and leave draft answers.

AI: I read 3 tasks.

1. "The page numbers in the PDF table of contents are wrong" — The "Creating PDFs" chapter of the user guide has the solution.

2. "I'd like to add more users" — This is a license inquiry, so the person in charge needs to check it.

3. ...

I've posted a draft answer with a link to the guide on task 1.

AI reads the email, finds supporting content in the documents, and writes a draft answer. The person in charge checks and polishes the draft, then just clicks “Send as Email” on the task screen to reply to the customer. People always make the decision to send an email.

A draft answer becomes visible to channel members as soon as it's posted. To keep it hidden from others until you've checked it, ask AI to leave it as a memo. Memos can't be sent as email, so post the polished content as an answer and then send it.

Ask AI to manage tasks, too

Case 4. Turn problems you find while working into tasks before you forget

While working, you run into problems you can't fix right now but mustn't forget.

User: Read the channel list. From now on, work in the "ManualWorks 6.0 Development" channel.

AI: I'll work in the "ManualWorks 6.0 Development" channel.

User: Create a task for the extraction problem we just found, and link it to the current task as a related task.

AI: I've created the task "Extracting archives in XML document import doesn't handle directory entries and parent paths."

I wrote down the symptom, cause, direction for the fix, and related files, and linked the two tasks as related tasks.

You get a task that holds all the context from the moment you found the problem. No more forgetting things you meant to “write down later.”

Case 5. Connect scattered tasks with relations

Large pieces of work are split into several tasks. Looking at the tasks alone, it's hard to tell what must be finished before the next thing can start.

User: Show me the related tasks of the following task. a000572a7b82af72

AI: This is the task "Make mail channels work without setup." It has no related tasks.

User: Task ddac6b485d8ff388 can start only after this task is finished. Link them so this task comes first, and link task 53c1d3692f6554f1 as a related task.

AI: I've linked the two tasks.

- "Document the X-MCP-Client header in the MCP setup guide": comes after this task

- "XML document import handles chapter and element property values differently from document properties": related task

There are three kinds of relations between tasks: related, preceding, and following. When AI reads a task, it can also see the IDs and titles of related tasks. If needed, it can read those tasks too, check the conclusions of earlier work, and then start the next job.

Case 6. Ask instead of clicking to change statuses and milestones

User: Remove the Opened category from this task and add the Closed category.

AI: I removed Opened and added Closed.

User: Add this release's milestone to the following task. 604caa9f5100766f

AI: I added the "6.0.25" milestone.

Just say the names of categories and milestones. AI reads the list and finds the right item, so you don't need to memorize IDs, and you can ask to add and remove in one request.

Case 7. “What should I do first today?” with AI

User: Show me the tasks in the "Assigned to Me" filter.

AI: There are 12 tasks.

1. Page numbers in the PDF table of contents are off

2. Hidden chapters show up in web viewer search results

...

User: Pick the ones that would be good to handle before the release.

AI: I recommend number 2 first, since it's small and affects customers. ...

It uses the same filters you use on screen. Beyond reading the list, you can discuss with AI what to handle first.

Case 8. Picked the wrong channel? Just say “Move it”

User: Move the task I just created to the "Customer Requests" channel.

AI: Done. The task type changed to match the channel.

It follows the same rules as moving on screen. For example, you must be the owner of both channels, the assignee must be a member of the target channel, and tasks received by email can't be moved.

It also records which AI did the work

Tasks and answers written with an AI tool have @mcp after the author's name, so you can tell them apart from posts written by people. If you add the X-MCP-Client header to your AI tool's settings, it records even which tool did the work, like John@claude-code or John@codex. This helps teams that use several AI tools when they look through the history.

Get started now

To start with MCP, just create an API key and register the server address in your AI tool. For setup, see Using ManualWorks in Claude Code/Codex/Gemini CLI, and for example requests for tasks, see Handling Tasks with MCP.

Hand one of today's assigned tasks to your AI tool. “Handle this task and leave an answer.” See for yourself how much lighter your day gets with this one sentence.