SkillAgentSearch skills...

OpenOctopus

Run AI agents on one server. Execute tools across every machine you own.

Install / Use

claude mcp add Zpoteiti -- npx -y github:Zpoteiti/OpenOctopus

If the server publishes to npm under a different name, use that package instead — check the repo README.

About this skill
🔌

MCP Server

Model Context Protocol server

Quality Score

81/100

Category

Automation

Supported Platforms

Claude Code
Claude Desktop

Our assessment of OpenOctopus

OpenOctopus scores 81/100 on our quality scale, 2453rd of 2,930 Automation skills we index.

Its MCP Server is 13 KB long, well organised into 17 sections with 13 code examples: a thorough specification that gives an agent plenty to work with.

It has 10 GitHub stars, so there is little community track record yet; judge it on its content.

Substance
30/30
Structure
20/20
Description
12/15
Adoption
4/20
Freshness
15/15

Maintenance, license and trust

  • The repository was last updated 9 days ago, so OpenOctopus is actively maintained.
  • It is released under the MIT license, a permissive license that allows use, modification and commercial use with attribution.
  • Its trust signals score 92/100, with 1 caution from licensing, adoption, age or documentation. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.

OpenOctopus compared with similar skills

All 4 of these similar skills score higher than OpenOctopus; compare them before choosing.

SkillScoreStarsUpdatedFormat
OpenOctopus (this skill)by Zpoteiti81109d agoMCP Server
Agent-Reachby Panniantong10097.4k6d agoCLAUDE.md
headroomby headroomlabs-ai10075.4ktodayCLAUDE.md
CowAgentby zhayujie10047.4ktodayCLAUDE.md
Scraplingby D4Vinci10087.7ktodayMCP Server

Frequently asked questions

How do I install OpenOctopus?
Run claude mcp add Zpoteiti -- npx -y github:Zpoteiti/OpenOctopus. The install tabs above show the steps for each supported agent.
Which AI agents does OpenOctopus work with?
It is written for Claude Code and Claude Desktop, as a MCP Server file. Other agents that read the same format can often use it too.
Is OpenOctopus safe to use?
It is MIT-licensed and scores 92/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
Is OpenOctopus still maintained?
The repository was last updated 9 days ago, so OpenOctopus is actively maintained.

OpenOctopus

Run one AI agent server and let it work in Server Workspaces or on paired Linux, macOS, and Windows computers.

OpenOctopus Server connects browser, Discord, and DingTalk channels with paired Linux, Windows, and macOS clients.

OpenOctopus is an alpha/demo release. Its Client can run commands and local MCP services with the permissions of your operating-system user. Read Security and current boundaries before exposing a Server or pairing a computer.

What it does

  • Serves a browser UI and API from the same Docker image.
  • Stores durable users, conversations, configuration, and MCP catalogs in PostgreSQL, with Server Workspace files in RustFS.
  • Provides personal and equal-permission shared Workspaces, file management, browser attachments, and transfers between the Server and paired computers.
  • Routes file tools and web_fetch to the Server or a selected Client.
  • Runs pipe commands and line-oriented PTY/ConPTY sessions on paired Clients.
  • Connects admin-managed Server MCP and per-device MCP over stdio, Streamable HTTP, or SSE, including tools, resources, templates, and prompts.
  • Connects one Discord Bot and one DingTalk Bot per user to the same durable conversation and Agent loop as the browser.
  • Uses native Pydantic AI Providers for Anthropic Messages, OpenAI-compatible Chat Completions, and OpenRouter.
  • Runs the Pydantic AI Harness with durable DBOS main and background child agents.
  • Ships six read-only built-in skill guides, available to every agent and in Workspace.
  • Runs scheduled jobs and Jev-gated Heartbeat checks. Dream reviews completed days' conversation text and updates the official Memory store, with change history and undo in Automations.

New accounts receive an editable SOUL.md file in their personal Workspace. The Memory page edits the same versioned PostgreSQL notes used by agents and Dream. Administrators can configure the default SOUL, Provider, Workspace quotas, Jev decision service, Server Web Fetch policy, users, and shared Server MCP from the UI.

See it in action

Server Workspace

Browse the Server Workspace alongside paired Clients, then manage files and agent instructions in one place.

OpenOctopus Workspace with personal files and an online paired Client.

Paired Clients

Pair a computer once, see whether it is online, and route the Agent to its Workspace, commands, and MCP services.

OpenOctopus Devices page showing an online paired Linux devbox.

External Channels

Connect a personal Discord or DingTalk Bot to the same durable conversation as the browser, with owner pairing and an explicit text-only allow list.

OpenOctopus Channels page for Discord and DingTalk configuration.

Architecture

Browser
  | REST + best-effort NDJSON stream
  v
OpenOctopus Server (one ASGI worker)
  |-- PostgreSQL: users, conversations, configuration, MCP catalogs
  |-- RustFS: Server Workspaces and uploaded attachments
  |-- Pydantic AI Harness + native model Providers + DBOS recovery
  |-- Jev decision API for Heartbeat Phase 1 and Dream
  |-- Discord Gateway + REST adapter
  |-- DingTalk Stream + OpenAPI adapter
  |-- Admin-installed Server MCP services with conversation-owned clients
  `-- Protocol v3 WebSocket
        `-- OpenOctopus Client
              |-- local file tools and transfers
              |-- exec: pipes + POSIX PTY / Windows ConPTY
              |-- Client web_fetch
              `-- Device MCP runtimes

The Server does not run Client shell commands. Device MCP and Client tools run on the paired computer. Admins install Server MCP once for all users. Each conversation connects to those services using its own client and refreshes the catalog during each tool-enabled model step. Idle conversation clients are retained for up to 10 minutes, subject to process-wide client and stdio limits. Server stdio services run as child processes with the Server OS user's permissions; HTTP/SSE clients connect to their configured remote endpoints.

Five-minute Docker quickstart

Requirements: Git, Docker, Docker Compose v2, and curl.

git clone https://github.com/Zpoteiti/OpenOctopus.git
cd OpenOctopus
cp .env.example .env

Edit .env and replace every replace-with-... value. In particular, use independent long random values for the PostgreSQL password, RustFS secret, JWT signing secret, and administrator registration token.

docker compose --env-file .env pull
docker compose --env-file .env up -d --wait
curl --fail http://127.0.0.1:8080/health

Compose starts PostgreSQL, RustFS, and OpenOctopus, creates the RustFS bucket, bootstraps the database schema, and preserves both data stores in named volumes. It listens on 127.0.0.1:8080 by default.

Open http://127.0.0.1:8080/register. Register the first administrator with the value of OPENOCTOPUS_ADMIN_TOKEN from .env. The same token can create additional administrators; omit it to create a regular member.

Configure the LLM Provider

Sign in as an administrator, open Admin settings, and configure:

  • API base URL: the unversioned origin, such as https://api.siliconflow.cn — do not append /v1;
  • API key;
  • Model, such as Qwen/Qwen3.5-4B;
  • the model context window and desired output/compaction limits.

Before saving, OpenOctopus calls GET {base_url}/v1/models and verifies that the configured model exists. Chat then uses the Provider's Anthropic-compatible /v1/messages endpoint. A Provider running on the Docker host must be addressed by an address reachable from the Server container; localhost inside the container refers to the container itself.

The administrator owns this shared Provider configuration and API key, and therefore bears Provider usage and cost for browser, channel, Cron, and Heartbeat Phase 2 turns and Dream memory proposals from every account on the deployment.

Configure Jev decisions

In Admin settings, set the separate Jev API base URL (for example, https://api.typesafe.ai, without /v1) and Jev API key. Saved keys are redacted; leaving the key blank retains it. Clearing the endpoint disables the configuration. Saving these settings requires no service request.

Use Check Jev to send a small explicit evaluation and display the observed status and check time. Opening settings reads the saved status. Endpoint or key changes reset that status to unchecked until a request succeeds. Heartbeat Phase 1 and Dream require Jev; missing configuration or failed/invalid decisions defer work. Selected Heartbeat tasks then execute through the normal LLM Agent. Cron follows its configured schedule.

The administrator owns Jev usage and cost for the deployment. The HTTP contract, failure handling, and recovery have mocked test coverage; live Jev authentication, deployment compatibility, latency/cost, and decision quality remain unverified.

Stop or remove the stack

Stop containers while preserving data:

docker compose --env-file .env down

Delete containers and all local PostgreSQL/RustFS data:

docker compose --env-file .env down --volumes

The second command is destructive.

To build the Server image from the current checkout instead of pulling it:

docker compose --env-file .env up -d --build --wait

Pair a Client

  1. In the browser, open Devices, create a device, and save the token shown once.
  2. Download the native one-folder bundle from GitHub Releases, or run the Client from source.
  3. Start it with the Server origin and device token.

Linux and macOS:

export OPENOCTOPUS_SERVER_URL='http://127.0.0.1:8080'
export OPENOCTOPUS_DEVICE_TOKEN='open…[redacted]'
./openoctopus-client/openoctopus-client run

Windows PowerShell:

$env:OPENOCTOPUS_SERVER_URL = 'http://127.0.0.1:8080'
$env:OPENOCTOPUS_DEVICE_TOKEN = 'open…[redacted]'
.\openoctopus-client\openoctopus-client.exe run

OPENOCTOPUS_SERVER_URL must be an HTTP(S) origin without a path, query, or fragment. Use HTTPS/WSS for a remote Server. See client/README.md for artifact names, source installation, lifecycle, and Client policy details.

Connect Discord or DingTalk

Open Channels in the browser and configure the Bot credential for Discord or the Client ID and Client Secret for DingTalk. Secrets are write-only. The DingTalk Client ID is also its robotCode; the current API verifies that identity but does not expose a platform Bot name or avatar, so OpenOctopus leaves those profile fields unknown instead of inventing platform metadata.

For Discord, enable Message Content Intent and grant the Bot View Channels, Read Message History, Send Messages, Send Messages in Threads, and Attach Files. The Channels page shows the same setup list.

The account owner proves their channel identity by sending the one-time pairing code to the Bot in a direct message. Other people are admitted only when the owner manually enters their exact platform user IDs in the allow list, one ID per line. An allow-listed non-owner gets text-only message access to the current conversation or the paired owner's direct message; there is no agent-to-agent channel protocol.

External channel history is visible in the browser as read-only history. The Server persists a complete Agent reply before the platform adapter splits it into bounded Discord or DingTalk messages. Each platform action is issued at most once and its outcome is stored; partial, failed, or unknown delivery is not retried automatically. Send a new message in the original channel to start a new Agent turn and try again.

Security and current boundaries

  • restrict_to_workspace=true confines OpenOctopus-resolved file paths and an exec session's initial working directory. It is an application path guard, not an operating-system sandbox, and it does not constrain shell commands, MCP, or networking.
  • Exec, PTY/ConPTY, and Device MCP inherit the permissions of the user running the Client. Install and pair them only on computers you trust.
  • Server and Client web_fetch have independent configurable denylists. Exec and MCP networking is open by design and can bypass those denylists.
  • Server and Device MCP environment variables and HTTP headers are stored in PostgreSQL as reversible plaintext and redacted from API responses. Remote MCP headers require an HTTPS endpoint; sending secret-bearing Device MCP configuration to a Client also requires WSS.
  • Attachments from allow-listed non-owners are rejected before any byte download, so they cannot reach the owner's Workspace or Client. Attachments sent by the paired owner may enter the owner's Workspace as authorized data; they are not executed automatically and OpenOctopus does not claim antivirus scanning.
  • The current Server uses one ASGI worker and process-local coordination. Do not run multiple workers or multiple Server replicas against one deployment.
  • The release bundles are unsigned, platform-native one-folder applications; they are not installers, services, or single static binaries.
  • PTY/ConPTY targets line-oriented REPLs and simple prompts. Full-screen TUI applications and reliable secret/password entry are outside the current support boundary.
  • Document conversion supports PDF, DOCX, XLSX, PPTX, and downloaded HTML. OCR, audio/video, archive recursion, and direct remote PDF/Office conversion are not enabled.
  • /health checks PostgreSQL and RustFS. Aggregate Server MCP conversation counts and sanitized errors are exposed through the administrator view.

For an Internet-facing deployment, keep OpenOctopus behind a TLS reverse proxy, use HTTPS/WSS, and

Truncated for display — read the full file on GitHub.

Related Skills

View on GitHub
GitHub Stars10
CategoryAutomation
Updated9d ago
Forks1

Languages

Python

Trust signals

92/100

From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.

1 low1 info