MCP Server

Overview

The Model Context Protocol (MCP) is a standard for connecting AI agents to external tools. Tasked implements MCP over stdio using JSON-RPC 2.0 with Content-Length framing (LSP-style).

This lets agents submit long-running DAG workflows, continue working on other tasks, and check back for results later. The agent never blocks waiting for a shell command to finish — Tasked handles execution, retries, and dependency resolution in the background.

Starting the MCP Server

tasked-server mcp --data-dir tasked-data

The server reads JSON-RPC messages from stdin and writes responses to stdout. Logs go to stderr so they don't interfere with the protocol.

Claude Desktop

Add this to your Claude Desktop configuration (claude_desktop_config.json):

{
  "mcpServers": {
    "tasked": {
      "command": "tasked-server",
      "args": ["mcp", "--data-dir", "tasked-data"]
    }
  }
}

Claude Code

Add to your project's .mcp.json:

{
  "mcpServers": {
    "tasked": {
      "command": "tasked-server",
      "args": ["mcp", "--data-dir", "tasked-data"]
    }
  }
}

Tools Reference

Tasked exposes eight MCP tools. Each tool auto-creates queues as needed, so agents can submit workflows without any setup.

tasked_submit_flow

Submit a DAG workflow for execution. Tasks run concurrently where dependencies allow. Auto-creates the queue if it doesn't exist.

Parameter	Type	Required	Description
queue	string	yes	Queue name (created automatically if it doesn't exist)
tasks	array	yes	Array of task definitions forming the DAG
webhooks	object	no	Optional { on_complete, on_failure } URLs for flow-level webhook notifications
queue_config	object	no	Queue configuration (only used when creating a new queue)

Each task in the tasks array accepts:

Field	Type	Required	Description
id	string	yes	Unique task identifier within the flow
executor	string	yes	Executor type: shell, http, or noop
config	object	no	Executor-specific configuration
input	any	no	Optional input data for the task
depends_on	string[]	no	Task IDs this task depends on
timeout_secs	integer	no	Timeout in seconds (default: queue default)
retries	integer	no	Max retry count (default: queue default)

The optional queue_config object accepts:

Field	Type	Default	Description
concurrency	integer	10	Max concurrent tasks
max_retries	integer	3	Default max retries
timeout_secs	integer	300	Default task timeout in seconds

Returns the flow ID, queue, state, and task count.

tasked_flow_status

Check the status of a submitted flow, including per-task state and outputs.

Parameter	Type	Required	Description
flow_id	string	yes	The flow ID returned by tasked_submit_flow

Returns the flow state, task counts, and an array of per-task statuses with their outputs and errors.

tasked_task_output

Get the output or error of a specific task within a flow.

Parameter	Type	Required	Description
flow_id	string	yes	The flow ID
task_id	string	yes	The task ID within the flow

Returns the task state, output JSON, and error message (if any).

tasked_list_flows

List active and recent flows, optionally filtered by state.

Parameter	Type	Required	Description
queue	string	yes	Filter by queue name
state	string	no	Filter by state: running, succeeded, failed, cancelled

Returns the queue name, count, and an array of flow summaries.

tasked_cancel_flow

Cancel a running flow and all its non-terminal tasks.

Parameter	Type	Required	Description
flow_id	string	yes	The flow ID to cancel

Returns a confirmation message on success.

tasked_create_schedule

Create a cron schedule that automatically submits a flow to a queue on a recurring basis.

Parameter	Type	Required	Description
queue	string	yes	Queue to submit flows to
cron	string	yes	Cron expression (5-field or 7-field with seconds)
flow	object	yes	FlowDef to submit on each trigger (same schema as tasked_submit_flow)

Returns the schedule ID, queue, cron expression, and creation timestamp.

tasked_list_schedules

List all schedules for a queue.

Parameter	Type	Required	Description
queue	string	yes	Queue to list schedules for

Returns an array of schedule summaries including ID, cron expression, and last/next fire times.

tasked_delete_schedule

Delete a schedule by ID.

Parameter	Type	Required	Description
schedule_id	string	yes	The schedule ID to delete

Returns a confirmation message on success.

Integration Examples

Agent workflow

A typical agent interaction with Tasked follows this pattern: submit a flow, continue working, then check back for results.

// 1. Agent submits a workflow
tasked_submit_flow({
  "queue": "agent-tasks",
  "tasks": [
    {
      "id": "fetch-data",
      "executor": "http",
      "config": {
        "url": "https://api.example.com/data",
        "method": "GET"
      }
    },
    {
      "id": "process",
      "executor": "shell",
      "config": { "command": "python3 process.py" },
      "depends_on": ["fetch-data"]
    }
  ]
})
// Returns: { "flow_id": "f_abc123", "state": "running", ... }

// 2. Agent checks status later
tasked_flow_status({ "flow_id": "f_abc123" })
// Returns: { "state": "succeeded", "tasks": [...] }

// 3. Agent retrieves specific task output
tasked_task_output({
  "flow_id": "f_abc123",
  "task_id": "process"
})
// Returns: { "state": "succeeded", "output": {...} }