- Home
- Skills
- Agents & Orchestration
- Cowork: Multi-Agent Task Inbox over MCP
Works with the AI Agents you already use
Cowork: Multi-Agent Task Inbox over MCP
Give your agent a shared task inbox. Dispatch, schedule and track work across Claude Code, Codex, Gemini CLI and local models through one self-hosted MCP hub.
$6
Cowork: Multi-Agent Task Inbox over MCP
Example session with this skill installed
Have Codex review src/server/ for queries that aren't tenant-scoped, and every weekday at 09:00 have the local Ollama model summarise yesterday's git log into CHANGELOG-draft.md.
- Read your context and instructions
- Compiled the cowork
Summary
Dispatched one review task to Codex now and scheduled a recurring changelog task on the local model. Both are in the shared Cowork inbox.
| # | Task | Brain | When | Status |
|---|---|---|---|---|
| 1 | Review src/server/ for tenant-scoping leaks | codex (pinned) | now | pending |
| 2 | Draft CHANGELOG from yesterday's git log | ollama-qwen (pinned) | Mon–Fri 09:00 (server time) | scheduled |
Calls made
get_dashboard() -> brains online: claude-opus, codex, ollama-qwen
register_agent(platform: "claude", agent_name: "laptop-session")
create_task(
title: "Review src/server/ for tenant-scoping leaks",
description: "Repo: ~/code/app. Check every DB query under src/server/ is filtered by orgId.
Deliverable: findings.md with file:line, the query, and a suggested fix.
Done = every file in src/server/ checked.",
from_platform: "claude", from_agent: "laptop-session",
context: { brain: "codex" }
)
create_task(
title: "Draft CHANGELOG from yesterday's commits",
description: "Repo: ~/code/app. Run git log --since=yesterday, group commits by feat/fix/chore,
write CHANGELOG-draft.md to the artifacts dir.",
from_platform: "claude", from_agent: "laptop-session",
context: { brain: "ollama-qwen" },
recurrence: { type: "weekly", weekdays: [1,2,3,4,5], atTime: "09:00" }
)
Each brief is self-contained because the receiving agent never sees this conversation. If Codex hits a rate limit, the hub moves task 1 to the next brain in its fallback chain.
Next steps
list_inbox(status: "done", limit: 5)later to read Codex's findings; the report is attached to the task card on the dashboard.- Watch the first scheduled run tomorrow at 09:00. Note that the time is the server's local time, so tell me if your hub runs in a different timezone.
Connects securely to your tools. The creator never sees your data.
About this skill
Cowork is a free, self-hosted MCP server with a web dashboard. Your coding agents share one task queue through it. This skill teaches your agent when to hand work to another agent and how to do it correctly. It covers registering with the hub, writing self-contained task briefs, pinning a task to a specific model ("brain"), scheduling one-off and recurring runs, asking a human for input mid-task, and reading results and artifacts back.
- Mixed vendors, one inbox: Claude Code, Codex, Gemini CLI/Antigravity, Hermes, Ollama.
- Fallback chains: when a model hits a rate limit, the task moves to the next brain.
- Scheduling: ISO date-times, or daily, weekly or cron recurrences.
- Human in the loop: form fields on the task card, and tasks park on
wait-input. - Plain files, no database. Docker image for amd64 and arm64.
Requires a running Cowork hub (MIT, free):
docker run -d -p 6868:6868 -v cowork-data:/data ghcr.io/slashman413/cowork:1.0.1.
The first start prints a generated API key. Setup and client wiring are in
references/setup.md.
Network behaviour (disclosed): the skill only calls the Cowork MCP server that you configure. It makes no other network calls, runs no scripts and collects no credentials. The API key stays in your MCP client config.
Requires a running Cowork MCP server (which the AI agent setups automatically after using this skill) and its optional API key if security is concerned. Agents that stop sending heartbeats are marked stale after about 10 minutes. Recurring schedules use the server's local time. Tasks are stored as plain files on the server, so keep credentials out of task descriptions.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
1 use
Times used to date
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 29 days ago
Requires
Permissions
The skill contains only markdown instructions. It runs no scripts and makes no network calls itself; your agent calls the Cowork MCP server that you configure. The hub API key stays in your MCP client config or the COWORK_API_KEY environment variable.