MCP Server Ready

Autonomous AI Developerson Your Infrastructure, on Any Model

mcptask.online - hand off a well-defined task in the evening, wake up to a merged PR and draft invoices. Your AI agents run on your machines against your real tests, on premium or cheap models - you pick.

dns
Model-agnostic, no lock-inRun Claude, GPT, Gemini or any open-weights model. Switch anytime. Pay for cheap inference, not premium APIs.
smart_toy
On your infrastructureAgents run on your machines, against your real database and your real tests. Your data never leaves your infra.
merge
Task in, merged PR + invoice draft outThe same lifecycle from backlog to Fakturoid / iDoklad invoice. No other tool covers the whole loop.
groups
Manage projects from chatManage your project from Claude Cowork, Claude Desktop or any MCP client. Same queue as the agents.

Start Free Trial See How It Works

For example, four runners, four different models - including cheap ones. They run on your machines against your real environment, not inside someone else's sandbox.

Sample view of a live runner activity dashboard

Runner activity

4connected·3working

sync_altAI runners stream their work live to this dashboard over the MCP protocol

#4821 Add OAuth login to the signup wizard

💭The callback already verifies state — reusing it for Google login…

Editapp/models/user.rb2.1s
Wire provider column into auth flow

#4805 Fix the flaky checkout system test

💬{"TASKRUNNER_RESULT": true, "status": "success", "pr_number": 1287}

#4818 Speed up the nightly import job

💬Tests are green — committing and opening a PR.

Bashgh pr create --fill4.0s
Create pull request

#4823 Rate-limit the public API endpoints

MCPProtocol

Live data streaming

auto_awesomeWhen a runner hits a failing test in code it was not asked to touch, it files its own tracked Bug task, branches off, fixes it, merges, and comes back. Nobody else ships that loop end-to-end.

peoplePerfect For

Built for Outcomes, Not Parity

Use cases written for the person signing the invoice - not the person writing the code

person_off

Capacity gap

Short on developers, can't hire fast enough

You have a backlog that won't shrink. The agents pick up the well-defined tasks overnight, log their work, and open PRs. Your team reviews in the morning instead of typing.

Problems Solved

check
Backlog grows faster than the team
check
Overtime costs and missed deadlines
check
Hard to find senior developers
check
Losing deals because of slow delivery

starAgents close PRs, humans stay in review

See Capacity Solutionsarrow_forward

savings

Cost ceiling

Bootstrapping - cheap models, not premium APIs

Premium model APIs eat your runway. mcptask.online is model-agnostic - the same agent harness runs Claude, GPT, or cheap open-weights models. Pay for what closes the task, not for the brand.

Problems Solved

check
Premium API bills spiral out of control
check
Forced to use a single vendor
check
No way to run cheaper models for the easy tasks
check
Runway shrinks while productivity grows

starSame harness, any model - including cheap ones

See Model-Agnostic Pricingarrow_forward

schedule

Async handoff

Need it done while you sleep

Hand off a well-defined task before you leave. Agents work overnight against your real tests. You wake up to a merged PR, a work log, and a draft invoice. No copy-paste of context the next morning.

Problems Solved

check
Loosing evenings to setup and context-switching
check
Time-zone waits for the offshore team
check
Manual work logging the next day
check
Invoices delayed by a week

starHand off in the evening, merged PR in the morning

See Async Workflowarrow_forward

forum

Chat-first management

Don't want to learn yet another PM tool

Manage projects from Claude Cowork, Claude Desktop, or any MCP client. Ask for a status, create a task, log a meeting - all from chat. The agents work from the same queue.

Problems Solved

check
Clicking through menus for every update
check
PM tools the team refuses to adopt
check
Reports that nobody reads
check
Context lost between tools

starWorks in Claude Cowork, Claude Desktop, and other MCP clients

See Chat-First Managementarrow_forward

The Challenge

Your AI Assistant Is Working Blind

AI coding tools are powerful, but they can't see your task system

content_copy

Copy-Paste Context Every Time

Every time you start with Claude Code or Copilot, you manually paste the task description, acceptance criteria, and context. Over and over. For every task.

timer_off

No Automatic Work Logging

Your AI assistant helped fix 5 bugs today. But your task system shows nothing. You spend 30 minutes reconstructing what was done and logging it manually.

playlist_remove

AI Can't Pick Up the Next Task

You want AI to work through your backlog automatically. But it can't see the queue, can't prioritize, can't move to the next task when done.

search_off

No Audit Trail of AI Work

Your manager asks: 'What exactly did the AI do on this feature?' You have no structured answer. Chat logs are scattered. Nothing is documented.

verifiedThe Solution

Give Your AI a Proper Task System

mcptask.online provides a dedicated MCP server that AI agents speak natively

install_desktop

Zero-Config Team Onboarding

One command (`bundle exec rake mcptask_runner:install`) provisions the MCP connection, Claude Code skills, baseline permissions, and your access token. On macOS it also offers a weekday 08:00 scheduler. `rake mcptask_runner:update` keeps skills current without overwriting local tweaks. Seniors stop wrestling with `.mcp.json` and token plumbing.

check_circleYour team is AI-ready in minutes, not hours

dns

AI Speaks MCP. So Do We.

Our MCP (Model Context Protocol) server lets AI assistants like Claude Code read tasks, understand context, and log work - using the same protocol they speak natively. No scraping. No fragile APIs. Clean, structured communication.

check_circleAI gets full task context instantly

smart_toy

Set It and Let It Work

One command (`bundle exec rake mcptask_runner:install`) provisions the MCP connection, Claude Code skills, baseline permissions, and your token. Then the AI agent fetches the next highest-priority task, completes the work, logs progress, and moves on. Human oversight when needed. Full automation when appropriate.

check_circleAI works through your backlog 24/7 — no manual plumbing

merge

Your Code Is Your Timesheet - and Your Invoice

Connect repositories via webhooks. Commits automatically log time. Pull requests link to tasks. Merges update status. On the Professional plan, the same data flows straight into Fakturoid or iDoklad as a draft invoice. No other tool covers the whole loop, from a task to a paid invoice.

Task -> Agent work -> PR -> Merge -> Timesheet -> Draft invoice in Fakturoid / iDoklad

check_circleOne lifecycle, from backlog to paid invoice

groups

Same Tasks, Different Interfaces

Humans use the elegant web interface. AI uses the MCP server. Both see the same tasks, same progress, same history. Complete transparency.

check_circleEveryone knows what's happening

account_tree

Epic > Story > Task > Subtask

Organize work the way it actually happens. Large features break into stories. Stories break into tasks. Tasks can have subtasks. AI and humans navigate the same structure.

check_circleReal projects, real organization

history

Who Did What, When

Every action is logged: task created by human, picked up by AI, progress logged, status changed, completed. Full attribution. Full history. Full accountability.

check_circleNever wonder 'who did this?'

rocket_launchQuick Start Guide

Four Steps to an Autonomous Developer

Create your account, reserve a machine for the agent, add the agent as a user, then run one install command

person_add

schedule2 minutes

Create Your Account

check_circleSign up with email (no credit card for trial)
check_circleCreate your workspace
check_circleImport existing tasks from Jira/Trello (optional)

computer

scheduleOne-time prep

Set Up a Machine for the Agent

check_circleDedicate a PC or Mac that will run the agent
check_circleCreate source-control accounts (GitHub/GitLab) so the agent can push branches and open PRs under its own identity
check_circleClone and run your project on that machine
check_circleInstall Claude Code

person_add

schedule1 minute

Add the Agent as a User

check_circleCreate a user for the AI agent in your mcptask.online project
check_circleThe agent picks up tasks, logs work, and changes status under its own identity
check_circleFull attribution: you always know who did what

terminal

schedule~2 minutes

Run the Installer — One Command

check_circleIn your Rails project run `bundle exec rake mcptask_runner:install`
check_circleIt writes `.mcp.json`, installs skills, merges permissions, and provisions your token
check_circleOn macOS it also offers a weekday 08:00 scheduler
check_circleFrom then on you just assign tasks — the agent does the rest

auto_awesome

After Setup

After setup, the agent works autonomously. You review completed work and steer priorities.

chat

terminalRead-only client

Just want to chat with your tasks?

If you only need a task-aware assistant in Claude Code or any MCP client, connect manually with this tiny `.mcp.json` snippet.

.mcp.json

{
  "mcpServers": {
    "mcptask-online": {
      "type": "sse",
      "url": "https://mcptask.online/mcp/sse",
      "headers": {
        "Authorization": "Bearer ${MCPTASK_TOKEN}"
      }
    }
  }
}

This is the lighter path — no runner, no autonomous PRs.

Ready to collaborate

attach_moneyPricing Plans

Simple Pricing That Scales

Start free. Upgrade when your AI does more work than you expected.

For Solo Developers with AI

Starter

$19/month

Prices exclude VAT (+ VAT where applicable)

Min. 2

Solo developers using AI assistants

check
2 users (1 human + 1 AI)
check
5 active projects
check
Unlimited tasks
check
MCP Server access
check
GitHub/GitLab integration
check
Basic reporting
check
Email support

Start Free Trialarrow_forward

Professional

$14/user/month

Prices exclude VAT (+ VAT where applicable)

Min. 5

Teams using AI-assisted development

check
Everything in Starter, plus:
check
Unlimited projects
check
Unlimited team members
check
Advanced MCP features
check
Sprint management
check
Advanced reporting
check
Team roles & permissions
check
Integrations (Jira, Trello, Redmine, Fakturoid, iDoklad)
check
Priority support
check
10 GB storage per user
check
API access

Start Free Trialarrow_forward

Dedicated Server

Enterprise

Min. 200

Organizations running multiple AI agents at scale

check
Everything in Professional, plus:
check
Dedicated server
check
Unlimited team members
check
Custom training
check
Continuous updates
check
Regular server maintenance
check
Dedicated server storage
check
Payment only for server operation and maintenance

Contact Salesarrow_forward

check_circleAll plans include unlimited AI agent connections via MCP server

See Detailed Pricingarrow_forward

format_quoteCustomer Stories

Teams Using AI Assistants Trust mcptask.online

format_quote

"Finally, my Claude Code sessions have full context. It reads the task, understands the requirements, and logs its work automatically. No more copy-paste ceremonies."

trending_up90% less context setup time

Martin V.

Full-Stack Developer

Using Claude Code daily

format_quote

"We run 3 coding agents overnight. mcptask.online is their task queue. Every morning we review what they completed. Full transparency, zero manual logging."

trending_up24/7 autonomous development

Sarah K.

Engineering Manager

AI-first startup

format_quote

"The GitHub webhook integration means I never log time manually. Commit message includes task reference, time is logged automatically. AI or human, doesn't matter."

trending_upZero manual time entry

Jan S.

Team Lead

10-person dev team

format_quote

"Switched from Jira. Setup took 30 minutes instead of 3 days. Now we're ready for AI assistants when we adopt them. Already cleaner than Jira was."

trending_up30 min setup vs 3 days

Petra D.

Project Manager

Digital agency

help_outlineFAQ

Frequently Asked Questions

How are you different from Copilot Workspace, Codex or Devin?

expand_more

Four things other AI coding tools don't have. First, model-agnostic: the same harness runs Claude, GPT, or any open-weights model - including cheap ones - so you never get locked into a premium API bill. Second, runs on your infrastructure against your real database and your real tests, not inside someone else's sandbox. Third, the only tool that covers the whole lifecycle, from a task in the backlog to a draft invoice in Fakturoid or iDoklad. Fourth, zero-config setup: one command provisions the MCP connection, skills, permissions, and token, while rivals leave your senior staff wiring agent configuration by hand.

What is MCP and why does it matter?

expand_more

MCP (Model Context Protocol) is the standard protocol that AI assistants like Claude use to communicate with external tools. mcptask.online provides a dedicated MCP server, meaning AI can read your tasks, update status, and log work using its native language - not through fragile web scraping or custom APIs. You don't build the connection manually - the runner provisions it during install.

Which AI assistants work with mcptask.online?

expand_more

Any AI assistant that supports MCP protocol: Claude Code, Claude Desktop, and other MCP-enabled tools. The MCP server provides read and write access to tasks, efforts, and project data.

Can AI agents work autonomously?

expand_more

Yes! Configure your AI to fetch the next highest-priority task, complete it, log progress, and move on. You set the boundaries (which projects, which task types). AI works within those boundaries.

How does GitHub/GitLab integration work?

expand_more

Add our webhook to your repository. When developers (human or AI) commit code with task references (e.g., '#Task-47 2h'), mcptask.online automatically logs the time and updates the task. Pull request merges can auto-complete tasks.

Is my data secure?

expand_more

Yes. All data encrypted at rest and in transit. Servers in EU data centers. GDPR compliant. Role-based access control means AI agents only see what you authorize.

Can I import from Jira or Trello?

expand_more

Yes! Professional and Enterprise plans include import tools for Jira and Trello. Your existing tasks, hierarchy, and history come over. Then connect AI and enhance your workflow.

Do I need technical skills to set up MCP?

expand_more

Practically no. `bundle exec rake mcptask_runner:install` runs the whole setup (MCP `.mcp.json`, token, skills, permissions, and a weekday scheduler on macOS) in about 2 minutes, and `rake mcptask_runner:update` keeps skills current without overwriting local tweaks.

What happens to my data if I cancel?

expand_more

Export all data (JSON, CSV) before canceling. We retain data 90 days in case you return. After 90 days, permanently deleted per GDPR requirements.

Q10

Do AI agents count as users for billing?

expand_more

Yes! Each AI agent has their own user in the system. AI agents can connect to MCP unlimited times, but their actions are logged.

Q11

Can I try before committing?

expand_more

Yes! 30-day free trial with full features. No credit card required. Connect your AI assistant during trial.

securityEnterprise Security

Enterprise-Grade Security for AI Workflows

shield

GDPR Compliant

Full EU data protection compliance

dns

EU Data Centers

Data stored in Frankfurt, Germany

hub

MCP Protocol

Native Model Context Protocol support

admin_panel_settings

Role-Based Access

AI sees only what you authorize

history

Audit Logging

Every AI action logged and traceable

speed

99.9% Uptime

SLA-backed reliability

verified_user

Your data is protected by enterprise-grade security

rocket_launch

Ready to Let AI Actually Help?

Hand off a well-defined task tonight. Wake up to a merged PR and a draft invoice.

calendar_today

30-day free trial - No credit card required

timer

One-command runner setup - Machine prep is the only one-time work

smart_toy

Unlimited AI agents - No extra charge for MCP connections

Start Free Trial Watch Videos on YouTube

verified_userTrusted by AI-forward teams. GDPR compliant. EU servers. Cancel anytime.

Autonomous AI Developerson Your Infrastructure, on Any Model

Runner activity

Built for Outcomes, Not Parity

Short on developers, can't hire fast enough

Bootstrapping - cheap models, not premium APIs

Need it done while you sleep

Don't want to learn yet another PM tool

Your AI Assistant Is Working Blind

Copy-Paste Context Every Time

No Automatic Work Logging

AI Can't Pick Up the Next Task

No Audit Trail of AI Work

Give Your AI a Proper Task System

Zero-Config Team Onboarding

AI Speaks MCP. So Do We.

Set It and Let It Work

Your Code Is Your Timesheet - and Your Invoice

Same Tasks, Different Interfaces

Epic > Story > Task > Subtask

Who Did What, When

Four Steps to an Autonomous Developer

Create Your Account

Set Up a Machine for the Agent

Add the Agent as a User

Run the Installer — One Command

After Setup

Just want to chat with your tasks?

Simple Pricing That Scales

Starter

Professional

Enterprise

Teams Using AI Assistants Trust mcptask.online

Trusted by AI-Forward Teams

Tasks Completed by AI

Hours Logged

MCP Server Uptime

Command to install

Frequently Asked Questions

How are you different from Copilot Workspace, Codex or Devin?

What is MCP and why does it matter?

Which AI assistants work with mcptask.online?

Can AI agents work autonomously?

How does GitHub/GitLab integration work?

Is my data secure?

Can I import from Jira or Trello?

Do I need technical skills to set up MCP?

What happens to my data if I cancel?

Do AI agents count as users for billing?

Can I try before committing?

Enterprise-Grade Security for AI Workflows

GDPR Compliant

EU Data Centers

MCP Protocol

Role-Based Access

Audit Logging

99.9% Uptime

Ready to Let AI Actually Help?

Product

Company

Legal

Follow us