Integrates with Brave Search API to enable web search capabilities for AI tasks.
Manages environment variables and API key configuration through .env files for secure credential storage.
Offers repository access for configuration examples and development contributions.
Provides access to Google's Gemini models (including gemini-2.5-pro) for AI task processing.
Handles package installation and management through npm for the MCP task server.
Allows tasks to utilize OpenAI's API and models like o3 and o3-pro for various AI capabilities.
Enables web search functionality through Perplexity's API for information retrieval during AI tasks.
Provides command execution capabilities via the run_command tool, allowing AI tasks to interact with the system.
@just-every/mcp-task
Async MCP server for running long-running AI tasks with real-time progress monitoring using @just-every/task.
Quick Start
1. Create or use an environment file
Option A: Create a new .llm.env
file in your home directory:
Option B: Use an existing .env
file (must use absolute path):
2. Install
Claude Code
Other MCP Clients
Add to your MCP configuration:
Available Tools
run_task
Start a long-running AI task asynchronously. Returns a task ID immediately (or batch ID for multiple models).
Parameters:
task
(required): The task prompt - what to performmodel
(optional): Model class or specific model name, or array of models for batch executioncontext
(optional): Background context for the taskoutput
(optional): The desired output/success statefiles
(optional): Array of file paths to include in the task contextread_only
(optional): When true, task runs in read-only mode (default: false)
Returns:
- Single task:
{ task_id, status, message }
- Batch execution:
{ batch_id, task_ids[], status, message }
check_task_status
Check the status of a running task with real-time progress updates.
Parameters:
task_id
(required): The task ID returned from run_task
Returns: Current status, progress summary, recent events, and tool calls
get_task_result
Get the final result of a completed task.
Parameters:
task_id
(required): The task ID returned from run_task
Returns: The complete output from the task
cancel_task
Cancel a pending or running task, or all tasks in a batch.
Parameters:
task_id
(optional): The task ID to cancelbatch_id
(optional): Cancel all tasks with this batch ID
Returns: Cancellation status and count of cancelled tasks
wait_for_task
Wait for a task or any task in a batch to complete, fail, or be cancelled.
Parameters:
task_id
(optional): Wait for this specific task to completebatch_id
(optional): Wait for any task in this batch to completetimeout_seconds
(optional): Maximum seconds to wait (default: 300, max: 600)return_all
(optional): For batch_id, return all completed tasks instead of just the first (default: false)
Returns: Task completion details with wait time, or timeout status
list_tasks
List all tasks with their current status.
Parameters:
status_filter
(optional): Filter by status (pending, running, completed, failed, cancelled)batch_id
(optional): Filter tasks by batch IDrecent_only
(optional): Only show tasks from the last 2 hours (default: false)
Returns: Task statistics and summaries with applied filters
MCP Prompts
The server provides MCP prompts that can be used to execute complex problem-solving strategies:
/solve
Prompt
Solves complicated problems by running multiple state-of-the-art LLMs in parallel and implementing their solutions.
Arguments:
problem
(required): The problem to solvecontext
(optional): Additional context about the problemfiles
(optional): Comma-separated list of file paths relevant to the problem
Strategy:
- Starts tasks with multiple models (grok-4, gemini-2.5-pro, o3, reasoning class)
- All tasks run in parallel to diagnose and propose solutions
- Tasks can create test files but cannot edit existing files
- First successful solution is implemented
- If a solution fails, retry with feedback to the same model
- Continues until problem is resolved
Example Workflow
Supported Models
Model Classes
reasoning
: Complex reasoning and analysisvision
: Image and visual processingstandard
: General purpose tasksmini
: Lightweight, fast responsesreasoning_mini
: Lightweight reasoningcode
: Code generation and analysiswriting
: Creative and professional writingsummary
: Text summarizationvision_mini
: Lightweight vision processinglong
: Long-form content generation
Popular Models
claude-opus-4
: Anthropic's most powerful modelgrok-4
: xAI's latest Grok modelgemini-2.5-pro
: Google's Gemini Proo3
,o3-pro
: OpenAI's o3 models- And any other model name supported by @just-every/ensemble
Integrated Tools
Task agents have access to a lightweight version of the tools available to Claude, optimized for autonomous task execution:
- Web Search: Search the web for information using
@just-every/search
- File Operations: Read and write files, with optional read-only mode
- Command Execution: Run shell commands (disabled in read-only mode)
- Code Analysis: Search and analyze codebases
Read-Only Mode
When read_only: true
is specified:
- Tasks can read files, search the web, and analyze data
- Tasks cannot modify files or execute commands that change system state
- Ideal for diagnostic tasks, code review, and solution planning
API Keys
The task runner requires API keys for the AI models you want to use. Add them to your .llm.env
file:
Getting API Keys
- Anthropic: console.anthropic.com
- OpenAI: platform.openai.com
- xAI (Grok): x.ai
- Google (Gemini): makersuite.google.com
- Brave Search: brave.com/search/api
- Perplexity: perplexity.ai
- OpenRouter: openrouter.ai
Task Lifecycle
- Pending: Task created and queued
- Running: Task is being executed with live progress via
taskStatus()
- Completed: Task finished successfully
- Failed: Task encountered an error
- Cancelled: Task was cancelled by user
Tasks are automatically cleaned up after 24 hours.
CLI Usage
The task runner can also be used directly from the command line:
Configuration
Task Timeout Settings
The server includes robust safety mechanisms to prevent tasks from getting stuck. All timeouts are configurable via environment variables:
Safety Features:
- Automatic timeout: Tasks exceeding
TASK_TIMEOUT
are automatically failed - Inactivity detection: Tasks with no activity for
TASK_STUCK_THRESHOLD
are marked as stuck - Health monitoring: Regular checks every
TASK_HEALTH_CHECK_INTERVAL
ensure tasks are progressing - Error recovery: Uncaught exceptions and promise rejections are handled gracefully
Development
Setup
Development Mode
Testing
Architecture
Contributing
Contributions are welcome! Please:
- Fork the repository
- Create a feature branch
- Add tests for new functionality
- Submit a pull request
Troubleshooting
MCP Server Shows "Failed" in Claude
If you see "task ✘ failed" in Claude, check these common issues:
- Missing API Keys: The most common issue is missing API keys. Check that your ENV_FILE is properly configured:
- Incorrect Installation Command: Make sure you're using
-e
for environment variables: - Path Issues: ENV_FILE must use absolute paths:
- Verify Installation: Check your MCP configuration:
- Debug Mode: For detailed error messages, run manually:
Task Not Progressing
- Check task status with
check_task_status
to see live progress - Look for error messages prefixed with "ERROR:" in the output
- Verify API keys are properly configured
Model Not Found
- Ensure model name is correctly spelled
- Check that required API keys are set for the model provider
- Popular models: claude-opus-4, grok-4, gemini-2.5-pro, o3
Task Cleanup
- Completed tasks are automatically cleaned up after 24 hours
- Use
list_tasks
to see all active and recent tasks - Cancel stuck tasks with
cancel_task
License
MIT
Author
Created by Just Every - Building powerful AI tools for developers.
remote-capable server
The server can be hosted and run remotely because it primarily relies on remote services or has no dependency on the local environment.
Async MCP server for running long-running AI tasks with real-time progress monitoring, enabling users to start, monitor, and manage complex AI workflows across multiple models.
Related MCP Servers
- AsecurityAlicenseAqualityA powerful MCP server that provides interactive user feedback and command execution capabilities for AI-assisted development, featuring a graphical interface with text and image support.Last updated -135MIT License
- AsecurityFlicenseAqualityAn intelligent MCP server that orchestrates multiple MCP servers with AI-enhanced workflow automation and production-ready context engine capabilities for codebase analysis.Last updated -3
- -securityAlicense-qualityAn enhanced MCP server that provides intelligent memory and task management for AI assistants, featuring semantic search, automatic task extraction, and knowledge graphs to help manage development workflows.Last updated -12MIT License
- -securityAlicense-qualityAn MCP server that lets agents and humans monitor and control long-running processes, reducing copy-pasting between AI tools and enabling multiple agents to interact with the same process outputs.Last updated -4MIT License