verifiedProof, not promise

AI wrote most of this company.

Karel Mráček, our autonomous AI developer, wrote the majority of the mcptask.online codebase — 1,200+ tasks, 2,000+ hours, merged PRs. This isn't an AI demo. This is AI that ships.

Start free trial

30 days free. MCP server included. No credit card.

architectureOur Difference

Our Product Builds Itself

While others demo AI features, we let AI do the work. Karel Mráček — our autonomous AI lead developer — wrote the majority of the mcptask.online codebase.

Karel Mráček

Autonomous AI Lead Developer

Karel runs on Claude Code and ships code, tests, and documentation around the clock. He is not a chatbot you copy-paste from — he is a member of the team with a backlog, time tracking, and merged pull requests.

smart_toy

1,200+

Tasks Completed by AI

schedule

2,000+

Hours Logged

cloud

HostingEU Data Centers

verified

Uptime99.9%

rocket_launchSee how the runner works

flagOur Mission

Make AI a Productive Team Member

AI assistants are powerful but isolated. They can't see your backlog. They can't track their own work. They work in chat windows, disconnected from your actual project management.

rocket_launchGive AI the tools it needs to be a real team member.

auto_mode

Unsupervised Execution

AI takes a task from assignment to merged PR and recovers when something breaks

cloud

Your Infrastructure

Runs against your real environment, your repositories, your data — not a sandbox

tune

Model-Agnostic

Use Claude, GPT, or local models that fit your budget and compliance needs

receipt_long

Full Lifecycle

From task to PR to timesheet to invoice — one continuous workflow

auto_storiesOur Story

Our Story

warning

error_outline

The Problem We Saw

In 2024, AI coding assistants became genuinely useful. Claude Code, GitHub Copilot, and others could actually write and fix code. But there was a problem:

AI couldn't see the task system.

Every session started the same way:

arrow_forwardCopy the task and its context from Jira into the AI chat
arrow_forwardAI does the work, human manually pastes the summary back into the task system
arrow_forwardManually log time spent — and repeat for every single task

Rinse and repeat. For every single task.

Meanwhile, AI assistants could communicate via MCP (Model Context Protocol) - a standard way to talk to external tools. But no task management system spoke MCP.

lightbulb

emoji_objects

The Eureka Moment

"What if the task system could speak MCP directly?"

We built an MCP server that connected to our existing task system (then called WorkVector). Suddenly:

check_circleAI could read tasks directly
check_circleAI could log its own work
check_circleAI could mark tasks complete
check_circleAI could pick up the next task automatically

Human time wasted on copy-paste: zero.

rocket_launch

trending_up

The Evolution

We realized we had something bigger than a feature. We had a new category:

task management for human-AI collaboration.

We rebuilt from the ground up:

starMCP server as a first-class feature
starWeb interface for humans (not abandoned)
starBoth sharing the same task pool
starFull transparency on who (human or AI) did what

mcptask.online was born.

directions_run

settings_input_component

Step 2: The Runner

An MCP server is the nervous system. The runner is the muscle. Once AI can read tasks, the next question is: can it finish one while you sleep?

check_circleRuns on YOUR machine against your real environment — real DB, system tests, screenshots
check_circleUses the model that fits your budget — from cheap local models to premium frontier models
check_circleWhen it hits an unrelated bug, it spawns its own task, fixes it, merges it, and returns to your work
check_circleWhen context overflows or it gets stuck, it recovers and finishes

auto_awesome

The runner doesn't just get stuck — it recovers. Hit an unrelated bug halfway through your task, it creates its own task, fixes it, merges it, and comes back. You wake up to two merged PRs.

Submit a well-defined task in the evening. Wake up to a merged PR and a logged timesheet.

rocket_launchExplore mcptask_runner

And the journey continues...

analyticsNumbers

Numbers

event

2024 (pivot from WorkVector)

Founded

trending_up

Small and focused

Team Size

smart_toy

1,200+

Tasks Completed by AI

schedule

2,000+

Hours Logged

verified_user

99.9%

MCP Server Uptime

location_on

Prague, Czech Republic (EU)

Location

terminalTechnology

For Developers: Why This Isn't Magic

code

BackendRuby on Rails 8.1 (latest)

web

FrontendHotwire Turbo (real-time, minimal JS)

storage

DatabasePostgreSQL

cloud

HostingAWS EU (Frankfurt)

dns

MCP ServerRuby (fast-mcp, SSE at /mcp/sse)

Why This Stack?

diamond

Rails

Mature, productive, ships fast. While others rewrite their stack every year, we ship features.

flash_on

Hotwire

Real-time updates without heavy JavaScript. When AI completes a task, you see it instantly.

storage

PostgreSQL

Reliable, powerful, handles complex queries for hierarchy and reporting.

security

EU Hosting

GDPR compliance, data sovereignty, fast for European teams.

Performance

speed

<2 seconds

Average page load

bolt

<100ms

MCP server response

sync_alt

<50ms latency

Real-time updates

verified

99.9% (last 12 months)

Uptime

psychologyCore Beliefs

Core Beliefs

groups

AI and Humans Together

The future isn't AI replacing humans or humans ignoring AI. It's collaboration. mcptask.online is built for both.

visibility

Transparency Over Magic

When AI does work, you should see exactly what it did. No black boxes. Full audit trail.

integration_instructions

Native Integration Over Hacks

AI speaks MCP. We speak MCP. No scraping, no fragile APIs, no middleware.

speed

Simple Over Complex

AI workflows should be easy to set up. 15 minutes to productivity, not 15 days.

bolt

Practical Over Theoretical

We build features people actually use. Every feature solves a real workflow problem.

favoriteOur Values

Our Values

smart_toy

AI-First, Human-Friendly

We design for AI integration first - but never at the expense of human experience. The web interface is clean, fast, and intuitive. AI features enhance human work, not replace human judgment.

shield

Privacy and Control

Your data is yours. AI agents see only what you authorize. Full audit logs. GDPR compliant. EU data centers. We don't train AI on your projects.

code

Predictable Capacity, No Hiring

The runner ships code, tests, and documentation steadily — no turnover, no sick days, no onboarding. You buy output, not headcount.

handshake

Honest Business

Transparent pricing. No hidden fees. Cancel anytime. Export your data anytime. We succeed when you succeed, not by locking you in.

groupsTeam

Small and Focused

We're a small team building a focused product. No massive sales department. No layers of management. Just builders solving problems.

Karel Mráček

Founder & Developer

Former agency owner, full-stack developer. Experienced the pain of disconnected tools firsthand. Built mcptask.online to solve his own problem - then realized others had the same one.

High-resolution portrét dostupný v tiskové sadě