Autonomous developer on your infrastructure.Any model — even a cheap one. Assign in the evening, review merged PRs in the morning.
The task queue is our mechanism, not our sales pitch.
Why the big players can't do this
Five things we do differently from generic AI agents.
Model-agnostic — no vendor lock-in
Our runner works with any model, including cheap and local ones. Copilot, Codex, and Claude Code lock you to their premium model. We don't.
Your real environment, not a sandbox
The agent works against your real database, system tests, and screenshots. Not a weak cloud sandbox.
Self-spawn: unrelated bug becomes its own task
Mid-task, the agent hits an unrelated bug → files its own tracked task → fixes it → merges → returns to the original. End-to-end.
Harness that finishes and recovers
The runner treats context overflow and unexpected errors as normal events. Pickup, retry, or fail-quietly — never stranded mid-task.
Effort flows straight into invoicing
Agent effort is logged as task minutes by default. Same mcptask.online that hosts the queue exports the timesheet to your invoice — no copy-paste.
Why this matters
Two recurring problems that derail AI agent programs before they deliver value.
No standardised task queue
Current workaround
Custom scripts, JSON files, manual assignment
Problem
Fragile, no visibility, does not scale
Black-box operation
Current state
Chat logs, scattered files, no structured data
Impact
No confidence in agent work
How teams use it
Illustrative scenarios — see the technical implementation for each one below.
Names and numbers are illustrative. Ask us for a guided demo for real figures.
I need a developer
Hire an autonomous agent for a specific epic. It works through tasks overnight while your team sleeps.
Technical implementation
Overnight bug fixer
Low-cost startup
Agents are free — you pay for users only. A cheap model + our runner = a per-task cost competitors can't match.
Technical implementation
Feature implementation team
I need it done overnight
Set the priority in the evening, review the PR in the morning. No night shifts, no overtime.
Technical implementation
Conservative autonomous mode
Feature Highlights
Smart Task Queue
Agent calls get_next_task() and receives:
- Highest priority available task
- Automatically locked (no other agent can pick it)
- Full context (description, acceptance criteria, related tasks)
- Time estimate (if available)
Priority Factors
| Priority level | Urgent > High > Normal > Low |
| Due date | sooner = higher priority |
| Task type | bugs before features, configurable |
| Age | older tasks surface |
Scoping Options
- Limit to specific projects
- Limit to specific task types
- Exclude certain labels
- Time-based restrictions
Full Activity Logging
Every Action Recorded:
- Task fetched by Agent-1 at 02:34:15
- Agent-1 logged 45 min of work at 03:19:22
- Agent-1 completed task at 03:45:08
- Agent-1 fetched next task at 03:45:12
Morning Dashboard Shows:
- Tasks completed overnight
- Total hours logged
- Success/failure rate
- Any blocked tasks
- Agent activity timeline
- Agent effort flows straight into invoicing
Human Oversight Controls
Safety Rails:
Control Types
| Task Types | Agents can only work on approved task types |
| Approval Gates | Certain tasks require human approval before agent starts |
| Daily Quota | Cap on how many tasks the runner will work autonomously per day |
| LaunchAgent (macOS) | Schedule the runner to fire on weekdays at 08:00 via a macOS LaunchAgent |
| Stop the Runner | The runner is just a rake task — Ctrl-C, kill the process, or disable the LaunchAgent |
Alerts
- PR opened (review in GitHub)
- Agent encountered blocker (visible in effort log)
- Agent failed task (visible in effort log)
- Real-time webhook notifications — coming soon
Multi-Agent Coordination
How It Works:
- Each agent has unique API key
- Task locking prevents conflicts
- Work distributed automatically
- No race conditions
Agent Fleet Management
- See all active agents
- Monitor individual progress
- Pause/resume specific agents
- Reassign work between agents
Scaling
Run 1 agent or 100. Same interface. Same pricing (agents are free).
For developers
The technical proof: one MCP server entry, a few rake commands, and a clear workflow.
Self-spawn, end-to-end
When the agent hits an unrelated bug mid-task, it does not stop and wait for you. Here is what actually happens:
Task #47 in progress
Agent-1 is working on the OAuth login refactor.
Hits unrelated bug
Agent-1 spots a flaky test that has nothing to do with OAuth.
Files task #53
Agent-1 creates a tracked task, labels it 'found by agent', links it back to #47.
Fixes #53
Agent-1 switches to #53, fixes the flaky test, opens a PR.
Merges #53
CI passes, auto-squash merges the PR.
Returns to #47
Agent-1 picks #47 back up from where it left off and finishes the original work.
All six steps happen without you. The runner picks up #47 again on its next loop if the agent crashes mid-task.
mcptask.online as Your Agent's Task Queue
The Agent Workflow
Agent calls get_next_task() via MCP
mcptask.online returns highest priority task
Agent works on task (coding, testing, etc.)
Agent calls log_effort() with progress
Agent calls complete_task() when done
Loop: Agent calls get_next_task() again
Task locking prevents duplicate work
Priority-based task selection
Automatic effort logging
Clear completion tracking
Continuous operation loop
Agent Configuration Examples
Three real mcptask_runner workflows. Run `bundle exec rake mcptask_runner:install` once, then schedule the command that fits your team.
These snippets show the actual rake commands you run or schedule. mcptask.online does not parse them — your host project and the runner do.
The real, copy-pasteable config
What every agent actually loads — one MCP server entry, one env var, one config file:
{
"mcpServers": {
"mcptask-online": {
"type": "sse",
"url": "https://mcptask.online/mcp/sse",
"headers": { "Authorization": "Bearer ${MCPTASK_TOKEN}" }
}
}
}
Overnight Bug Fixer
Let the runner work through today's highest-priority tasks overnight. Bugs naturally rank first because mcptask.online prioritizes urgent and high items.
# Run on a spare Mac/VM after installing the runner
command: bundle exec rake mcptask_runner:auto:squash:today
schedule: Mon–Fri 22:00 via macOS LaunchAgent (set by installer)
behavior: opens a PR per task and auto-merges after CI passes
quota: stops when the daily time budget is reached
Feature Implementation Team
Dedicate one host per feature epic. Use auto-squash for safe, self-contained Stories, or manual Story mode when every PR needs human review.
# One host per epic; the runner picks the next open task in the Story
auto_squash_feature:
command: bundle exec rake mcptask_runner:auto:squash:story[12345]
human_review_feature:
command: bundle exec rake mcptask_runner:manual:story[12345]
Conservative Autonomous Mode
No auto-merge. The runner executes one task at a time and leaves every PR open for review before it continues.
# One task, then stop; run again when you have reviewed the PR
single_review:
command: bundle exec rake mcptask_runner:manual:once
# Or loop through today's queue, leaving all PRs open
daily_review:
command: bundle exec rake mcptask_runner:manual:today
Monitoring & Safety
What you see and how you stay in control.
Agent Status Panel
- Active agents and current task
- Idle agents
- Paused agents
- Error states
Activity Timeline
- Task started by Agent-X
- Task completed by Agent-Y
- Effort logged
- Errors encountered
Metrics
- Tasks completed (24h / 7d / 30d)
- Hours logged by agents
- Success rate
- Average task duration
- Throughput trends
Scoped Access
- Agents only see authorized projects
- Agents only work on approved task types
- Granular permission control
Audit Trail
- Every agent action logged
- Full history exportable
- Clear attribution
Human Gates
- Every change goes through a pull request
- CI must be green before auto-merge
- No direct pushes to main
- Revoke MCPTASK_TOKEN to lock out an agent instantly
Alerts
- Critical errors and blocked agents
- Unusual patterns
- PR opened (review in GitHub)
- Real-time webhook notifications — coming soon
How Teams Use It
We run our coding agents through the queue. Every morning we review what they completed. We get more done without adding headcount.
Illustrative scenario
Composite based on user feedback
Not a real testimonial
Names and metrics are illustrative. For real numbers, ask us for a guided demo.
Tasks/week
Roughly doubles
depends on your queue
Human oversight time
Down significantly
varies by task type
PRs landed
Each one reviewable
before any merge
Cost per task
Lower than in-house
agents are free, you pay for users
Pricing for Agent Operations
MCP access for AI agents are free on all plans. You pay for account users only.
Starter
$19
/month
Best for: Solo operators with 1-2 agents
- MCP server access
- Unlimited AI agents
- Basic monitoring
- 1 human user
Professional
$14
/user/month
Best for: Teams running agent fleets
- Advanced MCP features
- Multi-agent support
- Agent dashboards
- AI work approval workflow
- Advanced reporting
- Priority support
Enterprise
Custom
Best for: Large-scale agent operations
- Dedicated MCP server
- Custom rate limits
- Advanced agent monitoring
- Fleet management tools
- SLA guarantee
- 24/7 support
Frequently Asked Questions
How many agents can I run?
Unlimited agents on all plans. Each agent needs its own API key for attribution.
Can agents conflict with each other?
No. Task locking prevents multiple agents from working on the same task.
What happens if an agent crashes?
Task is automatically unlocked after timeout. Another agent (or the same one after restart) can pick it up.
Can I restrict what agents can do?
Yes. Scope by project, task type, labels. Set approval gates. Limit actions (read-only vs. write).
How do I monitor agents overnight?
Activity log captures everything. Review in morning dashboard. Set alerts for critical events.
Is there an emergency stop?
Yes. Pause individual agents or all agents from dashboard. Immediate effect.
Run Your AI Agents with Confidence
Proper task queue. Full visibility. Safety controls. Start your free trial.
Unlimited agents. Full audit trail. Emergency stop included. 30-day trial.
Getting Started
Total beginner? You only need three things.
Install the runner
Add the mcptask_runner gem to your Rails project, run the installer, and you are connected in about two minutes.
See the full install walkthroughGet your MCPTASK_TOKEN
In mcptask.online, open Settings → API Access and generate a personal access token. The installer will prompt you to paste it during setup.
Open mcptask.onlineTell the AI which project to work on
Add a short mcptask.online section to your project's CLAUDE.md with your project_relative_id and account_code. Without it, the agent has no idea where to read or write tasks.
Jargon, in plain English
- MCP (Model Context Protocol)
- The open standard AI clients use to talk to external tools. Think of it as a USB cable between Claude Code and mcptask.online.
- Runner
- The mcptask_runner gem + a daily LaunchAgent (macOS) that fetches one task, lets your AI work it, opens a PR, and repeats.
- Agent
- Your connected AI client — Claude Code, Claude Desktop, or any MCP-compatible assistant. Agents are free on every plan.
- Quota
- The daily cap on how many tasks the runner will work autonomously. Stops the loop, no surprises.
- LaunchAgent
- macOS-only scheduler. The installer registers a LaunchAgent that fires the runner on weekdays at 08:00. On Linux/Windows, use cron, systemd, or Task Scheduler.