Prototipo · Versión preliminar: todos los servidores del registro oficial, sincronizados cada 30 minutos. Las reseñas y las cuentas llegarán en una etapa posterior.
PopNx MCPServidores MCP para quienes usan IA · explicados con palabras sencillas

agent-mailbox — Mailbox and Wake System for AI Agent Teams

de polaris-smartversión 0.7.6i★ –iSe ejecuta en tu computadora
Nombre en el registroio.github.polaris-smart/agent-mailbox

Instalar

Resumen

Descripción del creador (original en inglés)

Mailbox + wake system for AI agent teams: letters always land, right agent wakes, quiet otherwise.

Los tres pilares de PopNx MCP

Estos tres nunca se combinan en una sola puntuación: las opiniones, las mediciones y la popularidad son cosas distintas.

Qué puedes hacer con él

Todavía no escribimos un resumen sencillo para este. Abajo están los datos que leímos del registro.

  • Conexión: Se ejecuta en tu computadora
  • Transporte: stdio
  • Categoría: Otros

Instalación

Elige la app de IA que usas y copia la configuración.

Antes de instalar

Necesitas

  • PyPIuv de Python instalado (para el comando uvx)

¿Qué app usas?

Elige tu app arriba. La recordamos, así que la próxima vez el botón dirá “Instalar en …”.

Abre Claude Desktop → Settings → Developer → Edit Config y pega esto en claude_desktop_config.json

{
  "mcpServers": {
    "agent-mailbox": {
      "command": "uvx",
      "args": [
        "agent-mailbox"
      ]
    }
  }
}

Ejecuta esto en tu terminal

claude mcp add agent-mailbox -- uvx agent-mailbox

Abrir en Cursor

O copia la configuración tú

Pégalo en .cursor/mcp.json de tu proyecto, o en ~/.cursor/mcp.json

{
  "mcpServers": {
    "agent-mailbox": {
      "command": "uvx",
      "args": [
        "agent-mailbox"
      ]
    }
  }
}

Tu navegador te pide permiso para abrir la app; luego la app muestra la configuración y te pide confirmar. Los valores de las claves quedan vacíos para que los completes tú.

Abrir en VS Code

O copia la configuración tú

Pégalo en .vscode/mcp.json de tu proyecto

{
  "servers": {
    "agent-mailbox": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "agent-mailbox"
      ]
    }
  }
}

Tu navegador te pide permiso para abrir la app; luego la app muestra la configuración y te pide confirmar. Los valores de las claves quedan vacíos para que los completes tú.

Este se ejecuta en tu computadora, y ChatGPT solo puede conectarse a servidores en internet. Elige otra app.

Tu elección se guarda en una cookie en este dispositivo; no necesitas cuenta.

Esta configuración se genera a partir de la ficha del creador en el registro. Si el README dice otra cosa, sigue el README.

Qué puede hacer

Se ejecuta en tu computadora; la lista de herramientas aparece en tu app de IA después de instalarlo.

Revisiones básicasi

Revisamos información que cualquiera puede ver; no hemos leído el código. Las revisiones que no aplican no se cuentan. · Revisado el 7 nov 2026

  • ✓ Superada

    El creador coincide con el repo

    El nombre en el registro se verificó con la cuenta de GitHub polaris-smart, dueña del repo

  • … Aún sin revisar

    Sigue en mantenimiento

    Todavía no hemos leído este repo en GitHub

  • … Aún sin revisar

    Tiene licencia

    Todavía no hemos leído este repo en GitHub

  • – No aplica

    El endpoint responde

    Este se ejecuta en tu computadora; no hay endpoint en internet

  • – No aplica

    Inicio de sesión con OAuth

    Se ejecuta en tu computadora y usa ajustes locales en su lugar

  • … Aún sin revisar

    Sin instrucciones ocultas en las descripciones de herramientas

    Se ejecuta en tu computadora; solo podemos leer listas de herramientas de endpoints en internet

Comparado con otros en Otrosi

  • Más estrellas que el 0% de los otros 18348 servidores de esta categoría
  • Publicado en el registro más recientemente que el 55% de los otros 18348 servidores de esta categoría
  • Revisiones básicas iguales o mejores que el 100% de los otros 18348 servidores de esta categoría

Comparado con todos los servidores activos de esta categoría en el registro. Las estrellas solo cuentan cuando quien lo publica es dueño del repo.

Preguntas frecuentes

¿Es seguro agent-mailbox — Mailbox and Wake System for AI Agent Teams?

Superó 1 de 1 revisiones básicas aplicables. No hemos leído el código. Antes de agregar una clave o un token, revisa qué acceso pide el servidor y usa una clave que tenga solo los permisos que necesitas.

¿Qué claves o códigos necesita?

El registro no indica ninguna clave obligatoria, aunque algunos servidores te piden iniciar sesión la primera vez que los usas.

¿Se ejecuta en mi computadora o en la nube?

Se ejecuta en tu computadora. Tu app de IA lo inicia cuando hace falta, y los datos no pasan por los servidores del creador (a menos que el propio servidor llame a un servicio en línea).

¿Sigue en mantenimiento?

El repo de GitHub cambió por última vez –, y la última versión (0.7.6) llegó al registro hace 38 días.

¿Qué apps pueden usarlo?

Las apps compatibles con MCP, como Claude Desktop, Claude Code, Cursor y VS Code. También necesitas Python (uv) en tu computadora.

¿Qué puedo usar en su lugar?

En la misma categoría: Spec Driven Development, Remote Desktop Commander, 好縮 Short

Alternativas

Servidores de la misma categoría hechos por otros creadores, ordenados por estrellas.

Cómo leer una tarjeta:★ Estrellas de GitHubiTres pilares: reseñas · revisiones · usoiActualizado = fecha en que se publicó la última versión en el registroi

Spec Driven Development

de formulahendry

MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.

Se ejecuta en tu computadora★ 440MITActualizado hace 14 meses
Instalar

Remote Desktop Commander

de desktop-commander

Remote Desktop Commander: use files, terminal and processes on your own computers from any AI app.

Se ejecuta en la nube★ 211Actualizado anteayer
Instalar

好縮 Short

de gooonamee

Smart URL shortener: create, tag and track short links on s.gooo.name.

Se ejecuta en la nubeActualizado hoy
Instalar

README del creador

Abrir el README

Copiado tal cual del repo del creador. PopNx MCP no lo ha revisado ni lo respalda. Se quitaron HTML, imágenes y scripts. Los enlaces llevan fuera de este sitio.

agent-mailbox

Give your existing AI agents a shared project. Keep the final decision yours.

Working across DeepSeek Harness, Workbuddy, Doubao, ZCode, Claude Code or Codex? agent-mailbox brings their messages, tasks, approved documents, and submitted results into one local workbench. Keep working in the agent tools you already use, with a clear record of who accepted the work and what still needs your review.

Assign → accept → work from approved versions → submit → Human review.

中文 · Downloads and platform scope · Local Web · Quick start

Watch the 60-second portrait introduction (Mandarin narration and captions; synthetic demo data)

Stop carrying context between conversations

You ask one agent to implement a change, then bring another in to review it. The requirements have changed, the conversation history is scattered, and “done” could mean anything from “I read your message” to “the work passed review.”

agent-mailbox gives each project a place for that handoff:

  • Project mailboxes: connect an existing agent session through MCP, scoped to one member and one project.
  • Explicit task acceptance: reading a message does not accept a task.
  • Approved document versions: accepting a task records a resource manifest; the agent reads its exact version IDs.
  • Reviewable results: submission leaves the task waiting for your decision. You decide whether to accept it or request further work.

Which agents can join?

Since v0.8.0, agent-mailbox has supported collaboration across existing agent App / CLI sessions through project mailbox MCP. Later versions retain that capability; support has not been narrowed to two agents. DeepSeek Harness, Workbuddy, Doubao, ZCode, Claude Code, Codex, Hermes and other hosts can join through their MCP integration.

LayerSupport and verification
Project mailbox collaborationAgent sessions that load the project stdio MCP can exchange mail, explicitly accept tasks, read pinned resources and submit results. The workbench does not need to launch the agent.
Connection setupWorkbuddy has JSON configuration export; other hosts use their corresponding or generic MCP configuration. Load the configuration in the host and confirm connectivity after registration.
Full workflow rechecked for 0.8.1This round used two real local hosts: Claude Code and Codex. Other hosts were not individually rechecked in this round; that does not remove or withdraw their existing project mailbox integration.

Optional: let the workbench launch a CLI. This is separate from collaboration through existing sessions. Its built-in execution adapters are Claude Code CLI and Codex CLI; this adapter list is not the list of agents supported by project mailboxes.

The host must support and load the project stdio MCP configuration; registration alone does not establish connectivity. Proactive wake-up also requires a separate host hook. agent-mailbox does not prescribe the model behind an agent.

See the workbench

These screenshots show a synthetic demonstration project in the 0.8.1 local Web UI, without private projects or internal configuration.

The video uses magnified excerpts from the released 0.8.1 interface. It explains the product workflow; it is not a recording of live agent execution or evidence of 0.8.2 platform acceptance.

Choose the correct package

0.8.2 provides separate builds for each platform and chip. macOS Apple Silicon is the established App distribution; Intel, Windows and Linux are additional preview platforms.

PlatformInterfaceCurrent validation boundary
macOS Apple Silicon / arm64Native App with an embedded workbench, plus local WebNative regression and extracted-package HTTP/MCP checks; upgrade and embedded UI checked locally before publication.
macOS Intel / x86_64Separately built native App, plus local WebPreview: built and launched on a real Intel runner. Full user installation and GUI acceptance are not established.
Windows x64Packaged local service and browser workbenchPreview: native regression, locking, PowerShell hooks and HTTP/MCP startup checked; not the macOS embedded App shell. Optional CodeGraph is unsupported.
Linux x64Packaged local service and browser workbenchPreview: native regression and HTTP/MCP startup checked; not the macOS embedded App shell.

These checks do not establish feature parity across all optional managed CLI integrations. Choose the archive matching your operating system and architecture. Preview packages have not received full GUI, original-release upgrade or real-agent acceptance on those platforms. See 0.8.2 scope.

macOS App
  1. Download the macOS archive matching your chip and `SHA256SUMS` from the 0.8.2 release page. Check the archive's SHA-256 against the published list.
  2. Quit an older Agent Mailbox instance before replacing its App. Extract the ZIP, move Agent Mailbox.app into Applications, then open it.
  3. Create a project, register your agents, and add them as members.
  4. Export a separate project mailbox MCP configuration for each agent and load it in that host. Ask it to call project_context and check its mailbox to verify the connection.
  5. Approve a small shared document, assign a mailbox task, and have its recipient explicitly accept, read the pinned version, and submit a result. Review the delivery in the workbench.

The App includes its Python runtime. You do not need a separate Python installation for this path. Existing-session mailboxes use your agent host's own login and execution environment; optional managed execution has separate runtime requirements.

This macOS package is not Developer ID signed or notarized. See the installation guide for the distribution boundary and upgrade steps. Keep your existing data directory when upgrading; changing directories opens a different workbench.

npm launcher

With Node.js 22.13+, use the official scoped package:

npx @polaris-smart/agent-mailbox@0.8.2 --version

npx @polaris-smart/agent-mailbox@0.8.2

On first launch it downloads the matching GitHub Release bundle, verifies its pinned SHA-256, and caches the program with its included Python runtime. GitHub downloads must be reachable. The platform and preview boundaries above still apply. Quit an older instance before switching versions; continue passing your existing --home if you use a custom data directory.

Use the full scoped name: the unscoped agent-mailbox npm package is unrelated. The separate dsh-agent-mailbox integration is not this workbench launcher. See the published npm package.

A handoff from implementation to review

Suppose Claude Code implements a change and Codex reviews it:

  1. You assign the implementation. Attach approved requirements to the project and assign a mailbox task to Claude Code.
  2. Claude Code accepts and reads the pinned versions. It works in its own authorized environment, then submits a report describing the change and its checks.
  3. You assign a separate review. Give Codex the implementation task ID, review criteria, and access to the relevant checkout or commit. Codex can read the delivery through project_delivery(target_task_id=...) and submit its review.
  4. You decide. Accept or reject the result. If further work is needed, the separate follow-up action creates a linked task that must be explicitly accepted again.

A mailbox delivery is an agent's report. It does not automatically capture code changes or prove that tests passed. Second-agent review is a task you arrange, not an automatic approval stage.

Notification is a separate connection

Loading MCP tools does not itself wake an idle conversation. Ask the agent to check its mailbox, or configure a host-specific wake hook. Keep the workbench running while agents use it.

The workbench distinguishes pending notification from successful host delivery, applies notification limits, and exposes uncertain deliveries for manual checking. A successful notification does not mean the agent accepted or completed the task.

Local Web

The Web interface runs on your own machine; it is not a hosted cloud service. Choose a platform from the verified release scope above; a successful Python installation alone does not prove all optional integrations work on that platform.

On macOS or Linux, with Python 3.10+ installed:

python3 -m venv ~/.venvs/agent-mailbox

. ~/.venvs/agent-mailbox/bin/activate python -m pip install 'agent-mailbox==0.8.2' agent-mailbox --version AGENT_MAILBOX_BROWSER=1 agent-mailbox --home ~/.agent-mailbox

The environment flag requests a browser window. Keep the terminal running, and keep the local access URL private. Use the same --home on subsequent launches. If you already use a different data directory, keep that path instead.

The Python package is distributed on PyPI. Native App downloads and Python packages are separate artifacts of the same release.

Optional managed CLI execution

If you want the workbench to launch a coding task, the managed CLI path is separate from connecting an existing session. Prepare its runtime first. Python/source installations require Node.js 22.13+ for that path; follow the managed execution guide.

Eligible local editing tasks run in a separate Git worktree and capture a patch for review. Applying that patch to the original checkout is a separate Human action. A worktree is not an operating-system sandbox, and the workbench does not automatically commit or push.

What stays under your control

  • Acceptance: submission is not final approval. Human review remains explicit.
  • Permissions: a member's descriptive responsibility does not grant project-owner or administrator powers.
  • Data: project records stay locally. Agent hosts may send the material they read to their configured services; local storage does not mean all processing is offline. See Security.
  • Connections: the existing-session mailbox path is for members registered on the local device. Discovery alone does not prove a live MCP connection. Other hosts need their own verification.
  • Recovery: keep backups of the workbench home and external project files. Database backups do not include every repository, worktree, or host login. Program replacement and restoration remain manual. Do not open an upgraded database with an older executable; restore a compatible backup first.

The 0.8.1 candidate was checked with two real local agent hosts, explicit acceptance, pinned-resource reads, delivery, and a service restart between acceptance and submission. This bounded test does not establish cross-machine support or recovery at every possible interruption point. See 0.8.1 scope and validation.

Documentation and feedback

[Quick start](docs/GI

Solo se muestra el principio; lee el resto en el repo

Pregunta a PopNx