For freelancers and small businesses
MCP task manager
Your AI coding agent plans the app as a project, reports on each task as it works, and you steer it from the same place.
An MCP task manager lets your AI coding agent plan and report its work in a project you can see and steer. PrimeTask runs an MCP server on your computer, so Codex, Claude Code, Cursor or a local model can turn your plan into tasks, update them as it works and write down what it decided.
Updated · Checked with PrimeTask 1.0.5 · PrimeTask Team
PrimeTask
How it fits together
Select a part, in the drawing or here, to see what it does and what it feeds.
What flows
Every call, logged
Your AI agent to MCP server
Builds it, then keeps it current
MCP server to Project
Your own next task
Tasks to Compass
What the agent changed
Tasks to Dashboard
Every connection, in words
- Milestones is part of Project.
- Tasks is part of Project.
- Comments is part of Tasks.
- Timer is part of Tasks.
- Your AI agent is linked to MCP server.
- MCP server is linked to Project.
- Tasks is linked to Milestones.
- Tasks is linked to Compass.
- Tasks is linked to Dashboard.
- Every call, logged: Your AI agent to MCP server.
- Builds it, then keeps it current: MCP server to Project.
- Your own next task: Tasks to Compass.
- What the agent changed: Tasks to Dashboard.
How the parts of an MCP task manager connect, from your AI agent to the project you both work from
The problem
An AI coding agent plans in a chat you'll close, or in a to-do file in the repository. A few sessions later it's hard to tell what's done, what was decided and why, and each new session starts by working that out again.
The plan belongs somewhere you both work from, and that outlasts the chat.
What it uses
Five parts, one connected system
Each card opens its docs article in a new tab, for the detail behind the steps.
Bring Your Own AI
An MCP server on your computer, for the AI agent you choose
Docs for Bring Your Own AI (opens in a new tab)External Integrations
What the agent can see and change, and the name it writes under
Docs for External Integrations (opens in a new tab)Task Comments
What the agent did and decided, kept on the task
Docs for Task Comments (opens in a new tab)Task Time Tracking
The agent's time on each task, from its own timer
Docs for Task Time Tracking (opens in a new tab)Step 1
Turn on PrimeTask's MCP server
MCP Server settings (opens in a new tab)Open Settings → External Integrations and turn on the MCP Server. Read the explainer that opens, and PrimeTask creates your API key.
Turn on Write so the agent can create and update tasks. Keep Space Access on Selected only and tick just the Space that holds this app's work.
Step 2
Connect your AI agent
Bring Your Own AI (opens in a new tab)Under Quick Setup, click your agent. ChatGPT + Codex walks you through Codex. Claude Code, Cursor and LM Studio each copy a ready config and say where to paste it. Any other client that supports MCP connects with the details in the docs.
Restart the agent and ask it to list PrimeTask's tools.
Important. Never paste the API key into a chat with the agent.
Step 3
Give the agent the tools it needs
Tool catalog and agent name (opens in a new tab)In the Tool Catalog, choose Projects, the set made for project planning: tasks, milestones, time and comments, without the CRM and canvas tools. A local model in LM Studio works better with fewer tools.
Then set the Agent name. Its comments carry that name, Agent unless you change it, so you can tell its entries from yours.
Step 4
Turn your plan into a project
Milestones (opens in a new tab)Give the agent your plan and ask it to build the project: a milestone for each phase, and a task for each piece of work, linked to its milestone, with priorities and dates. It appears in PrimeTask as it's made.
Then ask it to read the project back, to catch a misunderstanding straight away.
Step 5
Have it report as it works
Task comments (opens in a new tab)Put the rules in the file your agent reads at the start of every session, such as CLAUDE.md for Claude Code or AGENTS.md for Codex. The checklist below has them.
The agent then runs the timer on the task in hand, moves the task along, and writes decisions, progress and blockers in the task's comments. It can only change comments it wrote.
Step 6
Steer it from PrimeTask
Prompts and the audit log (opens in a new tab)Change a priority or a date, or answer its question in the task's comments. With the rules below, it reads the thread before it starts. Ask it to run the project_status prompt for the project's health, milestones and blockers.
Every call it makes is in the Audit Log, with whether it worked.
Step 7
Start your own day with Compass
How Compass picks tasks (opens in a new tab)Add your own tasks to the project, such as reviewing each milestone. Then press Cmd + J (Ctrl + J on Windows): Compass puts overdue work first, then what's due today, then what's already in progress, each with the reason it's there.
The checklist
The whole list
Rules to paste into your agent's instructions file:
- Keep this app's work in the PrimeTask project named after it
- Before starting a task, read its comments
- Start the timer on the task you're working on, and stop it with a note when you finish
- Write each decision or blocker as a comment, marked Decision or Blocker
- When a task is done, complete it and say in a comment what should happen next
Variations
Make it yours
Four ways to change the setup to fit how you work.
Start read-only
Keep everything on your computer
Long runs while you're away
Your Dashboard
Reuse
Save it for next time
Set it up once, and start the next one from it.
A stand-up from the agent
daily_standup prompt: what was finished yesterday, what's planned today and what's blocked, gathered from PrimeTask.A Friday review
weekly_review prompt: progress by project, the tasks completed and the deadlines coming up.Questions
Common questions
Any that supports MCP, the Model Context Protocol: Codex, Claude Code, Cursor, Claude Desktop and LM Studio among them. PrimeTask doesn't subscribe you to an AI model or tie you to one vendor. You bring the agent you already use.
No. The MCP server runs on your computer and sends nothing anywhere. The agent you connect may send what it reads to its own provider, so a cloud agent's provider sees what the agent is given. A local model in LM Studio keeps it all on your machine.
No. The MCP server, with its task, project, milestone, time and comment tools, works on Standard. Pro adds the tools for PrimeCRM, custom fields and task dependencies.
No. It can read everything on a task, but it can only edit or delete comments it wrote itself. You can edit or delete anything in the thread, including its comments.
Every call it makes is in the Audit Log, with what it sent and whether it worked, so you can see exactly what changed. Keep Write off until you trust it, and make a backup before you turn it on.
Related
More setups like this
Other jobs, set up the same way.
Project management for one person
Every project you run alone, what matters today, and a clear run at doing it, without team software.
How to track how you spend your time
Every hour on the task it went to, from a timer, a focus session or time captured for you, with a week of totals to look back on.
How to track team tasks
One shared list where every task has an owner, a date and its discussion, synced through a folder your team already uses.
Set this up with your own work.
The free trial runs for 14 days and includes Pro.