MCP server de Loopback
Commit 12 nov 2026Proporciona una bandeja de entrada web con intervención humana para que los agentes de programación soliciten decisiones, revisiones y resúmenes a través de MCP.
También conocido como LoopBackTools/loopback
- Stars
- 0
- Herramientas
- 7
- Descargas, 30 días
- —
- Última versión
- —
- Último commit
- 12 nov 2026
- Licencia
- MIT
¿Qué es Loopback MCP?
Proporciona una bandeja de entrada web con intervención humana para que los agentes de programación soliciten decisiones, revisiones y resúmenes a través de MCP. Expone 7 herramientas, como ask_question, request_review, answer_follow_up.
Resumen generado a partir del README del proyecto
Acerca del MCP server de Loopback
Loopback permite ofrecer una bandeja de entrada web con intervención humana para agentes de programación. Permite que los agentes de IA soliciten decisiones, revisiones o resúmenes a través de una interfaz en vivo en lugar de quedarse bloqueados en una terminal o intentar adivinar las preferencias del usuario. Los desarrolladores utilizan esto para gestionar las interacciones de los agentes en múltiples proyectos desde un único panel centralizado.
Este MCP server se ejecuta en Cloudflare Workers utilizando Durable Objects y SQLite para garantizar que no se pierda ninguna solicitud. Se despliega en tu propia cuenta de Cloudflare y requiere un token secreto para la autenticación. Los usuarios pueden elegir entre un plan Free, que utiliza polling para clientes sin soporte de tareas en segundo plano, o un plan Paid para un polling más rápido y tiempos de espera más largos en preguntas bloqueantes.
El sistema incluye herramientas específicas como ask_question para consultas simples, request_review para inspeccionar código o diffs, y send_briefing para explicaciones complejas punto por punto. También cuenta con un plugin para Claude Code que redirige automáticamente las preguntas del agente a la bandeja de entrada.
Qué puedes hacer con él
- 1Revisión de código generado por agentesUn agente puede presentar un diff o un plan para tu aprobación antes de proceder con un cambio.
- 2Gestión de explicaciones complejasPuedes responder a resúmenes de varios puntos uno a la vez para resolver explicaciones técnicas complejas.
- 3Automatización de interacciones con Claude CodePuedes redirigir todas las preguntas de Claude Code directamente a tu bandeja de entrada web sin configuración manual.
Escrito a partir del README del proyecto. Cada caso de uso se verifica con una cita del mismo.
Herramientas de Loopback MCP 7
Detectado desde el READMEask_questionrequest_reviewanswer_follow_upsend_briefinganswer_briefingcreate_image_uploadopen_inbox
| ask_question | — | |
| request_review | — | |
| answer_follow_up | — | |
| send_briefing | — | |
| answer_briefing | — | |
| create_image_upload | — | |
| open_inbox | — |
Configuración
CLOUDFLARE_ACCOUNT_IDLOOPBACK_PLANLOOPBACK_URL
| Nombre | Descripción | Tipo |
|---|---|---|
| CLOUDFLARE_ACCOUNT_ID | — | |
| LOOPBACK_PLAN | — | |
| LOOPBACK_URL | — |
Verificaciones automáticas
Ejecutado 14 nov 2026Informe completo →Aún no verificado. Un servidor necesita una puntuación de 80, ningún hallazgo grave y suficientes datos para que podamos revisarlo. Cómo funcionan las verificaciones · Qué corregir para obtener más puntos
- Repositorio de origen vinculado y accesibleloopbacktools/loopback is public and reachablegithub.com/loopbacktools/loopback ↗Aprobado
- El paquete indica este repositorio como su origenNo verificado
- Procedencia de la compilación (atestación de Sigstore)No verificado
- Licencia de código abierto (OSI)MITspdx.org/licenses/MIT.html ↗Aprobado
- Commit o lanzamiento en el último añoLast commit or release 2026-11-12github.com/loopbacktools/loopback ↗Aprobado
- OpenSSF ScorecardNo verificado
- El nombre del paquete no es similar a un paquete popularNo verificado
- No hay avisos críticos en el paquete ni en sus dependenciasNo verificado
- Sin instrucciones ocultas para el modelo7 tool descriptions and the README read; no override phrases, credential paths or hidden charactersAprobado
- Las herramientas destructivas están etiquetadasNo tool named like delete, drop or execAprobado
- El endpoint remoto responde al handshake de MCPNo verificado
README
Ver en GitHubWhen an agent needs a decision, a preference or your sign-off, it asks through Loopback instead of guessing, or stopping in a terminal you are not watching. The ask appears in a live web inbox. You answer there, and your reply comes back to the agent as the result of its tool call, so it carries straight on.
Loopback is an MCP server on Cloudflare Workers: one Worker, one Durable Object per inbox, and the inbox itself served from the same deployment. You run it on your own Cloudflare account.
Features
- Questions and reviews. Agents ask questions, with or without choices, and show you work for a verdict: copy, a plan, code, a diff, a screenshot.
- Briefings, point by point. When an agent explains something that breaks into several points, you reply on each point on its own, drill into it through sub-points the agent adds, and resolve it. Each point is one line until opened, so what still needs you is plain to see.
- Ways to reply that fit the ask. Answer (pick choices, write your own words, or both), approve with an optional note, request changes, ask a question back, reply on a briefing's points, or dismiss.
- Agents don't stop for you. On clients that support MCP background tasks, an ask returns at once and the reply arrives when you give it. Other clients wait, kept alive by progress updates.
- One inbox for every project. Each ask names the project it comes from, so agents in many repositories share one inbox.
- Nothing lost. Asks and replies live in SQLite. A call retried after a dropped connection rejoins the ask it opened, and a reply waits for the agent that asked.
- A live inbox. Grouped by whose turn it is, updated as agents ask, with keyboard shortcuts and opt-in notifications.
- Claude Code plugin. Routes every question Claude Code has for you to Loopback, without your having to ask it to.
- Spec-first API. The REST API, event socket and MCP tools are defined once in TypeSpec, with a typed SDK for building your own inbox.
Tools
| Tool | What it does |
|---|---|
ask_question |
Asks you something, optionally with choices. |
request_review |
Shows you work for your verdict: approve, request changes, ask back or dismiss. |
answer_follow_up |
Answers a question you asked back, keeping the same ask open. |
send_briefing |
Explains something as points you reply on one at a time, and resolve. |
answer_briefing |
Answers your replies on a briefing, point by point, breaking points into sub-points. |
create_image_upload |
Hands out a one-time URL for an image to show in a review. |
open_inbox |
Gives the address of your inbox, or of one ask in it, so the agent can open it for you. |
The server tells agents when to use each tool, so asking an agent to "show me the copy" or "check with me before you pick" is enough.
Quick start
You need a Cloudflare account with R2 enabled, Node 24.22 or later (or 22.23 or later), and pnpm 11.
Loopback deploys with cf, Cloudflare's CLI, which is in
beta; pnpm install installs it for you.
git clone https://github.com/LoopBackTools/loopback.git
cd loopback
pnpm install
pnpm --filter loopback exec cf auth login
printf 'LOOPBACK_TOKEN=%s\n' "$(openssl rand -hex 32)" > ~/loopback.secrets
pnpm run deploy --secrets-file ~/loopback.secrets
The deploy builds the inbox, creates the Worker with your token as its secret, and provisions its R2 bucket.
Open the workers.dev URL the deploy printed, and sign in with the token from ~/loopback.secrets.
Keep the token somewhere safe, such as a password manager, then delete the file. It must be at least
32 characters, and it only ever travels in the Authorization header, which Cloudflare keeps out of
its logs.
Later deploys keep the token, so pnpm run deploy is enough. To change it, put a new one in a
secrets file and deploy with --secrets-file again.
Free or Paid
Loopback is built to run on Cloudflare's Workers Free plan, which holds each request to 10 ms of
CPU. Most of its requests take 2 to 9 ms. The first request a fresh isolate serves, and a blocking
question, take 20 to 65 ms: a copy held to 10 ms on Workers Paid let every one of them through,
but Free itself hasn't been measured. A deployment says which plan it is on with LOOPBACK_PLAN
when it is deployed:
LOOPBACK_PLAN=paid pnpm run deploy
Set it on every deploy: a deploy without it is for Free, and says so. The two differ only in what a deployment spends on agents waiting for you:
| Free (the default) | Paid | |
|---|---|---|
| A blocking question, from a client without background tasks | waits 10 minutes, then tells the agent to call again with the same arguments, which picks up the same question | waits until you answer |
| How often a background task is polled | every 5 s, then 15 s, then 30 s | every 2 s, then 10 s, then 30 s |
To check that a change fits Free, deploy a copy held to Free's 10 ms of CPU per request, drive it with an agent's traffic, and read back what each request cost:
pnpm --filter loopback run deploy --mode cpu-check --secrets-file ~/loopback.secrets
LOOPBACK_URL=https://loopback-cpu-check.<your-subdomain>.workers.dev LOOPBACK_TOKEN=... \
node apps/loopback/scripts/cpu-check.ts
CLOUDFLARE_ACCOUNT_ID=<your account id> \
pnpm --filter loopback exec node scripts/cpu-report.ts loopback-cpu-check <when the check began>
Delete the copy afterwards with pnpm --filter loopback exec cf workers delete loopback-cpu-check,
then empty and delete its bucket, loopback-cpu-check-files.
Connect an agent
Claude Code
Add Loopback's MCP server, for every project, with your token in LOOPBACK_TOKEN:
claude mcp add --scope user --transport http loopback https://<your-worker>/mcp \
--header "Authorization: Bearer $LOOPBACK_TOKEN"
Then install the plugin, which routes every question the agent has for you to Loopback without your having to ask it to. It holds only hooks, so it needs no URL or token:
claude plugin install loopback --marketplace LoopBackTools/loopback
integrations/claude-code explains what it does. Or let the
agent do it: every deployment serves /llms.txt, which tells an agent how to connect itself to that
deployment, for the way its agents sign in, and how to use Loopback once it has.
Any MCP client
Loopback serves MCP over Streamable HTTP at /mcp, authenticated with your token as a bearer
header. Most clients take a configuration like this:
{
"mcpServers": {
"loopback": {
"type": "http",
"url": "https://<your-worker>/mcp",
"headers": { "Authorization": "Bearer <your-token>" }
}
}
}
Behind a login, through the JWT gate, agents sign in with OAuth instead and need no token in their
configuration: docs/hosting.md explains it.
A client that supports MCP background tasks (io.modelcontextprotocol/tasks), such as Claude Code
2.1.304 and later, gets each ask back at once and carries on. A client without them waits for your
reply, which can take minutes. Loopback sends progress updates to keep the call alive, but a client
that caps tool calls at a fixed timeout needs that limit raised.
How it works
The Worker authenticates each request and hands it to a Space Durable Object, which holds the
asks, their threads and the connected inboxes in SQLite. Agents reach it through MCP; the inbox
through a REST API and an event socket. Images go to R2.
docs/architecture.md explains how it fits together.
Build your own inbox
The whole web app is published as @loopbacktools/app: compose it with your
own sign-in and workspaces, and it brings the inbox, its header, routing and the page on connecting
agents. The inbox alone is published as React components, @loopbacktools/inbox,
to render inside an app of your own: it reads and answers through the client you hand it, and moves
between asks through your routes. It is built only on @loopbacktools/sdk, a client for the REST
calls, the live event socket and the inbox state, with a promise API for any framework and an
Effect service, which packages/sdk explains. The API is also described
by an OpenAPI document, served at /api/v1/openapi.json, for clients in any language.
Hosting for many people
One deployment can serve many tenants. Point its JWT gate at any login that issues JWTs, and each
token's claims name the space it opens and the person; every space is its own Durable Object, and
agents sign in with OAuth. docs/hosting.md explains how.
Development
cp apps/loopback/.dev.vars.example apps/loopback/.dev.vars
pnpm dev # the Worker and the inbox, with hot reload
pnpm test # every package's tests
pnpm lint # oxlint
pnpm typecheck
pnpm format # oxfmt
pnpm generate # regenerate the API from its TypeSpec definition
| Path | What it is |
|---|---|
packages/protocol |
The API contract, in TypeSpec, and what is generated from it: types and the OpenAPI document. |
packages/server |
The server: Durable Object, HTTP API, event socket and MCP tools, written in Effect. |
packages/sdk |
A client for building your own inbox, in any framework. |
packages/inbox |
The web inbox as React components, built on the SDK. |
packages/app |
The web app around the inbox, for a deployment to compose with its own sign-in. |
apps/loopback |
The deployable Worker, and the app it serves, signed in with the deployment's token. |
integrations/claude-code |
The Claude Code plugin. |
Contributing
Issues and pull requests are welcome. CONTRIBUTING.md covers setting up, the
checks a change has to pass, and changing the API.
Security
Please report vulnerabilities privately, as SECURITY.md describes, not in a public
issue.
Licence
Preguntas sobre Loopback MCP
¿Qué herramientas proporciona Loopback MCP?
Expone 7 herramientas, incluidas ask_question, request_review, answer_follow_up, send_briefing, answer_briefing.
¿Cómo interactúan los agentes con Loopback?
Los agentes utilizan herramientas de MCP como ask_question, request_review y send_briefing para comunicarse con una bandeja de entrada web en vivo.
¿Puedo usar Loopback con Claude Code?
Sí, existe un plugin específico para Claude Code que redirige automáticamente cada pregunta que Claude Code tenga para ti hacia Loopback.