queue-mcp
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@queue-mcpAdd a task to fix the failing test"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
queue-mcp
A task queue for Claude Code. Line up work from another terminal, and Claude drains it one item at a time.
The problem
Type at the Claude Code prompt while a turn is running and your message gets steered into that turn at the next tool boundary. It does not wait. That is the right behaviour for "no, use the other file" mid-edit, but it makes it impossible to line up three follow-ups and walk away.
This queue lives outside the conversation, so nothing you add can disturb the running turn. Wanted upstream too: #50246 (217 reactions), #73661, #82348, #63190.
Related MCP server: taskqueue-mcp
Install
Via mcp-orchestrator, which also keeps it running:
mcp_install("queue")Or by hand:
python -m venv .venv
.\.venv\Scripts\python.exe -m pip install -r requirements.txt
.\run.ps1
claude mcp add --transport http --scope user queue http://127.0.0.1:8766/mcpTell Claude to use it
This step is what makes it automatic. Installing the server only provides
the queue; Claude will not use it for anything until told to. Add this to your
CLAUDE.md, then talk to Claude exactly as usual and every task you give it
goes through the queue.
## Task queue (queue-mcp)
The `queue` MCP server holds a per-session work queue. Drain it in order.
**Your session id** is the UUID in your scratchpad path. Pass it as
`session_id` to every queue tool. Never guess it, and never reuse one from an
earlier conversation.
**Everything the user asks you to do goes in the queue first, single tasks
included.** Do not weigh up whether a request is big enough to be worth
queueing. That judgment is exactly what fails: under momentum it always comes
out "this one is small, just start", and the rule quietly stops existing.
The test is **do vs say**, and it is the only judgment you make:
- If fulfilling the request means *doing* something, writing, editing, running,
building, investigating, it goes in the queue before any of it starts.
- If it means only *saying* something, a question about this conversation, an
opinion, a design you are discussing, answer inline and queue nothing.
On receiving a request that passes the test:
1. `queue_add` each step as its own item, in order, before doing any work. A
chain ("do X, then Y") is one item per step.
2. Check `queue_current`. If nothing is in progress, `queue_next` immediately
and start the first item.
3. If something is already in progress, just append and say so. Never jump the
queue or run two items at once.
**The loop:**
1. `queue_next` claims one item and marks it in progress.
2. Do the work.
3. `queue_complete` with a one-line result. Mark `failed: true` if it did not
work; do not silently mark it done.
4. Back to step 1. Stop when `queue_next` returns `drained: true`.
**Rules that matter:**
- **One at a time.** Never claim a new item while one is in progress.
- **`drained: true` means stop**, not "go look for more work".
- **A failed item does not block the queue.** Record it, report it, move on,
and flag it in the final summary.
- **The queue records the user's intent, not yours.** Never queue work they did
not ask for, beyond splitting their request into steps.
- **If the queue server is unreachable, say so and do the work anyway.** The
queue is a coordination aid, not a gate.Trim it to taste, but keep the do vs say test and the one-at-a-time rule. Those two carry most of the behaviour.
Adding work while it runs
.\q.ps1 add "fix the failing auth test"
.\q.ps1 add --next "urgent: revert the migration"
.\q.ps1 listClaude claims one item, works it, marks it done, then claims the next. It stops
when the queue reports drained.
Tools
Tool | Purpose |
| Append or insert an item |
| What is queued, what is running |
| Claim the next item |
| Mark done or failed |
| Edit the queue |
| Every known queue, and past work |
Worth knowing
Queues are per session. Every tool takes
session_id, because an MCP server cannot tell which session is calling it.Data lives in
~/.queue-mcp. Restarting the server loses nothing.The CLI writes the same files as the server, so adding work does not depend on the server being up.
The loop is model-mediated, not enforced. Claude queues and drains because
CLAUDE.mdtells it to, not because anything makes it. AStophook would make it strict.
Config: QUEUE_MCP_PORT, QUEUE_MCP_HOST, QUEUE_MCP_HOME.
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Related MCP Servers
- AlicenseAqualityDmaintenanceModel Context Protocol server for Task Management. This allows Claude Desktop (or any MCP client) to manage and execute tasks in a queue-based system.10198214MIT
- AlicenseAqualityCmaintenanceMCP server for "taming the Claude" with structured task queues.144670MIT
- Flicense-qualityDmaintenanceA task management MCP server that provides tools to create, list, complete, and delete tasks using pluggable storage backends. It enables users to interact with their task lists through natural language using MCP-compatible clients like Claude Desktop.
- Alicense-qualityDmaintenanceA task management MCP server that allows users to create, list, and organize tasks with priorities, labels, and projects using natural language. It provides a persistent SQLite-backed system for tracking due dates and managing task workflows directly within Claude Code.62MIT
Related MCP Connectors
A comprehensive Model Context Protocol (MCP) server that enables AI assistants to interact with yo…
Personal assistant MCP server with search, execute, packages, jobs, secrets, and integrations.
Hosted MCP server connecting claude.ai, ChatGPT and other AI apps to your own computer
Latest Blog Posts
- Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)By Om-Shree-0709 on .mcpAgent IdentityOAuth 2.1
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/mrconter1/queue-mcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server