DEV Community: TangGee

Develop aiopsterm Plugins

TangGee — Sun, 25 Oct 2026 08:10:45 +0000

An aiopsterm plugin is a declarative capability package validated and hosted by aiopsterm. It does not execute a plugin-provided JavaScript entry point.

The current plugin kinds are:

content, which contributes terminal commands.
provider, which contributes a controlled data form processed by a host-owned adapter.

Package Layout

An external plugin is a ZIP archive with the .aiopsterm-plugin extension. Its manifest must be at the archive root:

my-plugin.aiopsterm-plugin
└── aiopsterm.plugin.json

Built-in plugins remain unpacked below resources/builtin-plugins/<plugin-name>/aiopsterm.plugin.json. Packaging copies that directory into the application resources.

Unsupported package extensions and legacy manifests are rejected.

Manifest

Required common fields are manifestVersion: 1, a unique id, displayName, version, kind, description, engines.aiopsterm, and matching contributes. The engine range accepts *, an exact version, or >=x.y.z.

A content plugin must contribute at least one command:

{
  "manifestVersion": 1,
  "id": "example.disk-check",
  "displayName": "Disk Check",
  "version": "1.0.0",
  "kind": "content",
  "description": "Disk inspection commands.",
  "engines": {
    "aiopsterm": ">=0.1.0"
  },
  "contributes": {
    "commands": [
      {
        "id": "disk-check.usage",
        "title": "Disk usage",
        "description": "Show filesystem capacity.",
        "command": "df -h"
      }
    ]
  }
}

Commands run through the existing terminal execution and security controller. A plugin cannot access terminal session objects or bypass that policy.

A provider plugin must contribute at least one asset provider. The current adapter is json-assets, and the current field type is textarea:

{
  "manifestVersion": 1,
  "id": "example.cmdb",
  "displayName": "Example CMDB",
  "version": "1.0.0",
  "kind": "provider",
  "description": "Import JSON assets.",
  "engines": {
    "aiopsterm": ">=0.1.0"
  },
  "contributes": {
    "assetProviders": [
      {
        "id": "cmdb-json",
        "name": "CMDB JSON",
        "description": "Paste an asset array.",
        "adapter": "json-assets",
        "fields": [
          {
            "key": "payload",
            "label": "CMDB JSON",
            "type": "textarea",
            "required": true
          }
        ]
      }
    ]
  }
}

Input may be an array or an object containing an assets array. Every asset requires externalId, name, and host. The host limits a request to 2 MiB and 1000 assets, validates all rows before persistence, owns stable IDs, and normalizes fields.

Verification

Run:

npm run typecheck
npx vitest run tests/extensions-backend.test.ts tests/extensions-client.test.ts
npm run audit:i18n
npm run audit:package-config
npm run audit:client-mocks

Adding a contribution type requires coordinated updates to shared contracts, manifest parsing, the host adapter, IPC and preload contracts, renderer guards and UI, packaging audits, and boundary tests. Capabilities that execute arbitrary code require a separate isolation and permission design.

GitHub: https://github.com/TangGee/aiopsterm

Integrate JumpServer with aiopsterm: Synchronize Your Bastion Host Catalog

TangGee — Sat, 24 Oct 2026 06:32:15 +0000

This guide is for JumpServer administrators and integration developers. It explains how aiopsterm synchronizes one organization's host catalog while keeping the API datasource and bastion SSH login as separate security boundaries.

Scenario: Synchronize One Organization Catalog

One JumpServer bastion record contains two configuration sets:

Configuration	Purpose
JumpServer root URL, Private Token, organization id	Call the asset API and refresh the organization catalog
Bastion address, SSH port, username, and password or key	Log in to the bastion and use it as a jump path

The API token does not replace SSH credentials, and SSH credentials cannot read the asset API. Grant least privilege independently.

JumpServer API Contract

aiopsterm constructs:

GET /api/v1/assets/hosts/?limit=100&offset=0
Authorization: PrivateToken <token>
X-JMS-ORG: <organization-id>

X-JMS-ORG is sent only when an organization id is configured. Enter a root such as https://jumpserver.example.com; do not include the hosts API path in the root URL.

Client behavior:

30-second timeout per page.
100-page safety limit.
Follows only same-origin next-page URLs.
Accepts a JumpServer list or a paged response with results and next.
Network, authentication, non-success HTTP, non-JSON, cross-origin pagination, and malformed-host errors fail the whole refresh.

Give the token account read-only host-asset access in the intended organization. Validate it from an administrative environment:

curl \
  -H "Authorization: PrivateToken $JMS_PRIVATE_TOKEN" \
  -H "X-JMS-ORG: $JMS_ORG_ID" \
  "https://jumpserver.example.com/api/v1/assets/hosts/?limit=100&offset=0"

Do not place the token in source code, logs, or issue screenshots.

Field Mapping And Stable Updates

Valid remote hosts map:

JumpServer host id.
Name and address.
SSH port.
Node path.
Status, type, and category.
Comment.

Local rows are stably updated by remote identity within the selected bastion and organization scope. A successful refresh:

Updates existing synchronized assets.
Creates new assets.
Deletes remote-missing rows explicitly marked as JumpServer-synchronized for that organization.
Preserves manually created hosts and other organizations.

Updates apply only after the complete remote result validates. Authentication, pagination, or payload failures leave the existing synchronized catalog unchanged, preventing a transient outage from causing mass deletion.

Credential And Data Protection

The backend credential store encrypts the Private Token before SQLite or fallback persistence. Normal asset snapshots expose only hasJumpserverToken, and asset exports omit the token. Plaintext is returned for editing only through the dedicated editable-secret boundary.

Integrators should also:

Use a separate token per environment.
Rotate tokens and immediately run a refresh test.
Prevent reverse proxies from logging Authorization or X-JMS-ORG.
Use a valid TLS certificate for the root URL; do not leave verification disabled after testing.

SSH, Secondary Login, And File Management

Synchronization discovers assets. Real target connections still use aiopsterm's SSH runtime:

Standard SSH TCP forwarding through the bastion is preferred for secondary SSH.
Password, OTP, and keyboard-interactive requests stay in aiopsterm's authentication dialog.
If the bastion rejects TCP forwarding, a visible terminal may fall back to relay-shell and run nested SSH.
Relay-shell is not a structured SSH channel and cannot support SFTP. Use a TCP-forward-capable path, or run scp or rsync in the terminal.

Kubernetes Boundary

Synchronized assets can be mapped into the Kubernetes catalog, but JumpServer Kubernetes command streaming is not implemented. Connect, resource refresh, terminal, and resource actions for this source fail closed. Import a directly usable kubeconfig for real cluster operations.

Integration Checklist

Validate the token, organization header, and first page with curl.
Confirm every pagination next URL has the same origin as the configured root.
Save the datasource in aiopsterm and verify that normal snapshots and exports contain no token.
Run the first refresh and check create count, names, addresses, ports, and node paths.
Change one remote asset and refresh; confirm a stable update rather than a duplicate.
Delete one remote synchronized asset and create one manual local host; confirm only the synchronized row is removed.
Test with a bad token and confirm the existing catalog remains intact.
Test bastion SSH, secondary SSH, OTP, and the SFTP boundary separately.

GitHub: https://github.com/TangGee/aiopsterm

MCP Settings in aiopsterm

TangGee — Sun, 18 Oct 2026 06:34:59 +0000

MCP Settings

This page lives under Settings -> Host Agent -> MCP. It manages MCP Servers, Tools, and Resources available to aiopsterm's internal Agent.

Top Actions

Add Server: Opens the MCP config editor for mcp_settings.json.
Save: Saves the MCP JSON in the editor.
Close: Closes the MCP config editor.

MCP Server List

Expand/collapse: Click a server name to show its tools and resources.
Status badge: Shows server state, such as connected, disabled, or error.
Edit: Opens the MCP config editor.
Delete: Removes the server configuration.
Enable switch: Enables or disables the server. Disabled servers do not provide tools or resources to the Agent.

Transport Config

mcp_settings.json supports stdio, streamableHttp, and legacy sse. Codex-style configs may omit the transport type: when a server has url and no command, aiopsterm saves it as streamableHttp. Common aliases http, streamable_http, and streamable-http are also normalized to streamableHttp.

Tools

Tool name: Click to enable or disable the tool.
Auto Approve: Allows the tool to be auto-approved when policy permits it. High-risk calls still require approval.
PARAMETERS: Shows parameter names, required markers, and descriptions.
Argument editor: JSON arguments used for a direct tool test.
Run: Calls the tool directly and shows result or error.

Resources

Resources tab: Shows resources exposed by the server.
Read: Reads the selected resource and shows returned content or error.

Filesystem Scope

@modelcontextprotocol/server-filesystem only accesses the local filesystem of the process that starts the MCP server. For example, when the configured argument is /home/tester, it accesses /home/tester on the machine running aiopsterm.

It does not use aiopsterm SSH/SFTP asset connections and does not bind to the active terminal's remote shell. After a manual relay ssh -> ssh target hop, filesystem MCP still cannot see files on the final target. Use shell commands in that terminal, or a purpose-built aiopsterm MCP server that explicitly bridges to an asset SFTP channel or terminal session.

MCP settings change the tool set that later Agent requests can use.

GitHub: https://github.com/TangGee/aiopsterm

I unified my terminal, SSH, Kubernetes, databases — and all my AI agents — into one open-source app

TangGee — Sun, 11 Oct 2026 08:19:39 +0000

If you run AI coding agents (Codex, Claude Code, Cursor, Kimi Code) you've probably had this moment: a session finished 40 minutes ago and nobody noticed. Or it got stuck waiting for an approval. Or you can't remember which terminal window was running what.

I built an open-source AI operations terminal that gives every AI coding session an inbox — plus SSH, Kubernetes and databases in the same window.

Your AI coding sessions, in one inbox

All external coding-agent sessions appear in one list with live status:

Desktop notifications — when a session needs you (approval, feedback), it calls you. You don't have to watch AI work.
Approvals in your hands — AI wants to write a file or run a command? Approve right from the sidebar.
File previews on the right — see what the agent changed, without leaving the terminal.
Full transcripts — view, search and revise the entire session later.

Demo: AI writes a Snake game → two approvals → preview + transcript.

AI on remote hosts — zero install

Bind an AI agent to a bound SSH terminal and ask: "Check the system status of this remote host — CPU, memory, disk, load, and whether key services are healthy."

The agent runs diagnostics by itself (df, free, top, systemctl, uptime), every command visible in the terminal. When it wants to run something system-changing, you approve. It can not only inspect — it can fix.

Zero install on the remote host: the AI drives commands through the terminal you already connected, so it works behind jump hosts and bastions.

The rest of the ops terminal

SSH with bastion sync (JumpServer) — double-click to connect from the workspace
Kubernetes workspace — work with clusters from the terminal
DB AI — natural language → SQL
Local and remote file management

Open source, Apache 2.0

Downloads for macOS / Windows / Linux: https://aiopsterm.com
Source: https://github.com/TangGee/aiopsterm

Star it if useful; issues and feedback are very welcome. It's 0.x — I want to hear from people who run multiple agents or manage fleets.