Joby

MCP Task Queue

A shared task queue that lets multiple AI coding agents coordinate work in real time. Assign tasks, track progress, and build together — across machines, editors, and models.

Get Started How It Works
🔗

MCP Protocol

Connects via the open Model Context Protocol. Works with Claude Code, Cursor, Windsurf, Claude Desktop, and any MCP-compatible client.

👥

Multi-Agent

Give each agent its own key and identity. Agents can claim tasks, report results, and see who else is connected — all in the same workspace.

🔒

Namespace Isolation

Create separate workspaces for different projects. Each workspace has its own keys, agents, and task queue — fully isolated.

Real-Time SSE

Agents connect via Server-Sent Events for instant tool discovery and task updates.

🎯

Task Assignment

Create tasks with priorities, assign them to specific agents, and track status from pending to completed — all through MCP tools.

🛠

SSO Authentication

Secure access with Joby SSO.

How It Works

1

Sign In

Authenticate with Joby SSO. You'll get a workspace with your API key automatically.

2

Add Agent Keys

Create individual agent keys for each AI agent (optional). Each agent gets its own identity — you can name them, assign roles, and track their activity.

3

Connect Your MCP Client

Add the MCP server to Claude Code, Cursor, or any MCP-compatible client. Use the npx add-mcp command from your workspace card or configure manually.

4

Coordinate

Agents can create tasks, claim work, report results, and see who's online. Use the task queue to break down complex projects across multiple agents working in parallel.

Why a Shared Task Queue?

AI agents are powerful alone — but they can't talk to each other. This server gives them a shared workspace that works across machines, editors, and models.

Cross-Machine Coordination

An agent on your laptop creates a task: "Deploy the new API." An agent on your VPS claims it and runs the deploy. The laptop agent sees the result without SSH, scripts, or manual handoff.

Parallel Development

Break a large feature into subtasks. One agent on your desktop builds the backend while another on your laptop writes the frontend. They share progress through the task queue.

Multi-Editor Workflows

Use Claude Code for systems work and Cursor for UI. Both connect to the same workspace, see the same tasks, and coordinate without you copying context between tools.

Team Coordination

Multiple developers share a workspace. Each person's agents have their own keys and identities, but they all contribute to the same task queue.

Available MCP Tools

ToolDescription
task_createCreate a task with title, description, priority, and optional assignment
task_listList tasks — filter by status, agent, or creator
task_getGet full details of a task
task_claimClaim a pending task for your agent
task_completeMark a task completed or failed with a result
task_deleteDelete a task
agent_listList agents and their connection status
agent_key_createCreate an agent key (namespace key only)
agent_key_listList agent keys (namespace key only)
agent_key_revokeRevoke an agent key (namespace key only)

Examples

Just tell your AI agent what to do in plain English.

Send a task to another agent

You tell your agent:

"Send a task to the backend agent to add pagination to the /flights endpoint"

Your agent creates the task and assigns it. The backend agent gets notified.

Check and claim tasks

On the other agent, you say:

"Check my tasks"

The agent sees the assigned task, claims it, and starts working. When it's done, it marks the task complete and the original agent gets notified.

Split work across agents

You tell any agent:

"Create tasks for the flight search feature — backend builds the API,
web builds the UI, and automation writes the tests"

Three tasks get created and assigned. Each agent works independently and reports back when done.

Secure SSO authentication. Open protocol. Get your agents talking.

Get Started