mcpField
V
Vytrix

Vytrix

Local
github.comData & Analytics2Updated 1w ago

About Vytrix

Vytrix is a read-only system activity monitor for macOS and Linux, for people investigating host, application, project, and container activity. Its dependency-free Python collector supplies live or imported telemetry to a dashboard, with history and evidence views.

Key features

  1. 01Group processes by app or executable
  2. 02Monitor Docker and Podman containers
  3. 03Associate listening ports with projects
  4. 04View activity history and anomalies
  5. 05Import or export telemetry snapshots
  6. 06Collect host data with a Python file

Use cases

  1. 01Find which application is using CPU or memory
  2. 02Identify a process listening on a project port
  3. 03Inspect container resource use
  4. 04Review historical activity and anomalous ranges

MCP support

The project documents an MCP server that lets MCP clients query the collector through six read-only tools, with additional optional cluster or Fleet Pro tools, four prompts, and two resources.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

mcp.json · Universal config
{
"mcpServers": {
"zyvorai-vytrix": {
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"-e",
"VYTRIX_URL",
"-e",
"VYTRIX_TOKEN",
"ghcr.io/zyvorai/vytrix-mcp:0.8.0"
],
"env": {
"VYTRIX_URL": "<VYTRIX_URL>",
"VYTRIX_TOKEN": "<VYTRIX_TOKEN>"
}
}
}
}
Run commanddocker run -i --rm -e VYTRIX_URL -e VYTRIX_TOKEN ghcr.io/zyvorai/vytrix-mcp:0.8.0
Environment variables to set
VYTRIX_URLBase URL of the Vytrix collector, e.g. http://host.docker.internal:9847 or https://vytrix.example. Plain HTTP is accepted only for localhost; inside a container 127.0.0.1 is the container itself, so set this explicitly.
VYTRIX_TOKENSecretBearer token of the collector. Read from the environment only; never printed.

Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.

Packages
ocighcr.io/zyvorai/vytrix-mcp:0.8.0

Tools

No tool list yet

Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.

Details

From the project README

Zyvor Vytrix

verify License Node Python Platforms

Zyvor Vytrix: see what your machine is really doing

See what your machine is really doing.

A read-only system activity monitor for macOS and Linux. Applications grouped the way you think about them, processes, listening ports, project folders, and Docker and Podman containers, in a macOS 27 window, a macOS 26 Liquid Glass one, or an Adwaita one. A single-file Python collector with no dependencies feeds it.

Read-only · Dependency-free collector · macOS 27 + Liquid Glass + Adwaita · Docker and Podman aware · Local-first

🎮 Live demo · 🌐 Site · 🚀 Quickstart · 🧩 How it fits · 🔒 Security · 📖 API

Why Vytrix

When this happensVytrix gives you
Activity Monitor lists 24 "Google Chrome Helper" rows and you want one numberProcesses grouped by executable or .app bundle, with CPU and memory per application and a process sheet one click away
"Which container is eating the box, and is it Docker or Podman?"State, CPU, memory against its limit, network I/O and published ports for both runtimes, from read-only CLI calls
"What is listening on 5173?"Listening ports tied to the application and its project folder
You want a monitor on a remote server without an agent suiteOne Python file, one systemd unit, HTTPS and a bearer token
You will leave it running on a production hostIt is read-only: it never kills, signals or changes anything

See it

Try the live demo: the real dashboard, running in your browser on simulated telemetry (nothing to install, nothing leaves the page).

Download for Mac: the native app, v0.7.0, DMG and pkg (Apple silicon, macOS 26+); the release notes say whether the build is signed and notarized.

Vytrix: overview, an application's processes, containers, the Adwaita window style, then dark mode

About 12 seconds against the built-in simulated telemetry: overview, an application's processes, containers, switching to the Adwaita window style, then dark mode.

macOS 27, lightmacOS 27, dark
Overview, macOS 27, lightOverview, macOS 27, dark
Liquid Glass, lightLiquid Glass, dark
Overview, Liquid Glass, lightOverview, Liquid Glass, dark
Adwaita, lightAdwaita, dark
Applications, Adwaita, lightContainers, Adwaita, dark
Phone (390px)Phone, containers
Overview on a phoneContainers on a phone

Every image here is the app's built-in simulated telemetry (host zyvor-dev-01). A web page cannot read your processes, so the hosted preview says so on screen. See the UX contract.

How it fits together

A browser dashboard talks to a read-only collector over HTTPS; the collector reads the host

The dashboard starts on clearly labeled simulated telemetry. To see your own machine, run the collector and connect (live), or import a one-off snapshot (offline).

Mac cluster

Optional and read-only: a small worker on each Mac pushes its snapshot to a coordinator, and Monitor → Mac cluster shows every machine's CPU, memory, disk, network, applications and containers. There is no command channel. See docs/CLUSTER.md.

Native Mac app

A SwiftUI app (native/) that shows the same telemetry in a native macOS 27 window. It starts the bundled collector on loopback with a random token held in memory, or connects to one you run elsewhere (Connect…, HTTPS unless it is localhost). It is read-only, like everything else here.

Download Vytrix 0.7.0 for Mac (Apple silicon, macOS 26 or newer): Vytrix-0.7.0.pkg installs into Applications, or open Vytrix-0.7.0.dmg and drag Vytrix there. The v0.7.0 files are Developer ID signed and notarized. If you use a build that is not (a local or CI build without Apple secrets), the first launch is blocked: open System Settings → Privacy & Security and choose Open Anyway next to the Vytrix message. Each file has a .sha256; check it with shasum -a 256 -c Vytrix-0.7.0.pkg.sha256. v0.3.0 remains available as a hand-signed DMG. The app bundles the collector, which needs Python 3.10+ (see below).

Or build it from source:

./scripts/build-native.sh      # native/build/Vytrix.app, ad-hoc signed development build
open native/build/Vytrix.app   # or add --args --demo for simulated telemetry
./scripts/shots-native.sh      # docs/ux/native-*.png, from the simulated telemetry
./scripts/build-native-widgets.sh   # the same app plus the widget and App Intents (needs xcodegen), see docs/WIDGETS.md

Needs macOS 26 or newer, the Xcode command line tools, and Python 3.10+ for the collector (the system /usr/bin/python3 is 3.9; the app looks for Homebrew's first). Overview, Applications, Containers, Projects, Alerts and Settings are in, grouped Monitor, Workloads and System like the dashboard, along with a menu bar item (CPU, host totals, alert count, busiest apps). Closing the window keeps Vytrix running in the menu bar; Quit stops it and its collector. On the resource pages, anomalous ranges are shaded and a dragged range opens an evidence sheet (needs a collector, so not in demo mode). Alerts use the dashboard's rules (threshold default 80%, critical at +10) and post local notifications, once per alert, never for demo data. The CPU, Memory, Disk, Network and Battery pages and Settings (appearance, accent, threshold, snapshot import and export, data source) are in too. A System widget and two Shortcuts actions ("What is using my CPU?", "How is my Mac doing?") are in docs/WIDGETS.md. Built and run on macOS 27.2.

Overview, lightApplications, light
Native overview, lightNative applications, light
Containers, darkOverview, dark
Native containers, darkNative overview, dark

Quickstart

1. The dashboard, on simulated data (Node ≥ 22.13; npx [email protected] … works without installing pnpm):

git clone https://github.com/zyvorai/zyvor-vytrix.git && cd zyvor-vytrix
npx [email protected] install --frozen-lockfile
npx [email protected] dev            # http://localhost:5173

2. Your own readings, offline (Python 3.10+, Linux or macOS, no root):

python3 agent/vytrix.py --once > snapshot.json

In the dashboard choose Connect → Import collector snapshot (or Settings → Import). Nothing leaves your machine.

3. Live telemetry

export VYTRIX_TOKEN="$(python3 -c 'import secrets; print(secrets.token_urlsafe(32))')"   # keep it out of source control
python3 agent/vytrix.py --allow-origin http://localhost:5173

Enter http://127.0.0.1:9847/v1/snapshot and the token in Connect. The dashboard polls every two seconds. The token stays in tab memory: it is not stored in localStorage and not sent anywhere else. For another machine, serve HTTPS with --tls-cert/--tls-key (or a TLS reverse proxy that passes Authorization). The UI refuses plain-HTTP collector URLs except on loopback.

Collector flags
FlagDefaultPurpose
--oncePrint one snapshot, then exit
--bind, --port127.0.0.1, 9847Where to listen. Plain HTTP on a non-loopback address prints a warning
--allow-originnoneExact browser origin allowed by CORS. Repeatable
--tls-cert, --tls-keyPEM files: serve HTTPS (needed for remote browser access)
--ui-upstreamProxy non-API paths to a dashboard URL (same-origin deploys)
--interval2Seconds between samples
--database~/.local/share/vytrix/history.sqliteHistory (30-day retention, directory 0700, file 0600)
--no-containersSkip Docker and Podman
--container-interval5Seconds between container polls

Endpoints: GET /v1/snapshot, GET /v1/history?since=…&limit=… (both need the bearer token and an allowed origin), and unauthenticated GET /healthz. Everything is a read; mutations are not supported. See docs/API.md.

Views

SectionViews
MonitorOverview, CPU, Memory, Disk, Network, Battery
WorkloadsApplications (grouped), Containers (Docker and Podman), Projects (developer servers by folder)
SystemAlerts (threshold checks), Settings (window style, appearance, accent, reduced transparency, thresholds, import/export)

Also: search, CPU or memory sort, pause and resume, 1m/5m/1h activity history (captured while the tab is open, at most 1,800 samples), and JSON import/export.

Install as a service

macOS launchd, Linux systemd, remote Linux, Docker and Podman

scripts/install-macos.sh --allow-origin http://localhost:5173     # launchd agent, token in the login Keychain
scripts/install-linux.sh --allow-origin http://localhost:5173     # systemd user unit (--system for a hardened system unit)

Both accept --dry-run, --port and --no-containers.

Deploy to a Linux server

./scripts/deploy-remote.sh 203.0.113.10 deploy      # HOST USER, or user@host
./scripts/deploy-remote.sh HOST USER --quick        # sync + restart, no build
./scripts/deploy-remote.sh HOST USER --verify-only
./scripts/deploy-remote.sh HOST USER --dry-run      # prints the plan; no SSH

It syncs the checkout over SSH (one multiplexed connection) to ~/.vytrix/app, builds it there (with a private Node 22 if the system Node is older), and installs two systemd units: the dashboard on 127.0.0.1:8787 and the collector on https://HOST:30847 with a self-signed certificate that covers the host, proxying the dashboard on the same origin. The token is created once in ~/.vytrix/env (0600). Open the URL, accept the certificate, Connect (the endpoint is pre-filled) and paste the token. Needs SSH, passwordless sudo, Python 3.10+, curl, openssl, rsync. Details: docs/RELEASE.md.

Read-only by design

Vytrix reads processes, ports, resources and containers; it never kills, signals or changes anything

The dashboard has no login of its own: put it behind your own access control. What the collector token protects, and what it does not, is in SECURITY.md.

Verification

pnpm typecheck && pnpm lint
pnpm test          # telemetry schema tests + Python collector tests
pnpm build
pnpm test:e2e      # Playwright: WebKit, Chromium, mobile WebKit
pnpm shots         # regenerate docs/ux screenshots (simulated telemetry)

What was run where, and what was not, is listed honestly in docs/RELEASE.md and docs/TESTING.md.

Documentation map

I want to…Read
Call the collector, or learn what each metric meansdocs/API.md
Let an MCP client ask the collector (six read-only tools, optional cluster/Fleet Pro tools, four prompts and two resources; registry name io.github.zyvorai/vytrix, image ghcr.io/zyvorai/vytrix-mcp:0.8.0, both published from the v0.8.0 release)docs/MCP.md
Monitor several Macs (read-only)docs/CLUSTER.md
Fleet history, roles, alerts, Prometheus exportdocs/FLEET-PRO.md
Test on macOS and Linuxdocs/TESTING.md
Release or deploydocs/RELEASE.md
Understand the design rulesdocs/design/UX-CONTRACT.md
Report a vulnerabilitySECURITY.md
ContributeCONTRIBUTING.md, AGENTS.md
Regenerate screenshots, the demo GIF, social images and the Pages sitedocs/README.md
See what changedCHANGELOG.md

Scope

An initial read-only release. It does not kill processes, control fans or volume, inspect Bluetooth devices, measure per-app network or disk I/O, or collect GPU metrics. App memory is RSS, not unique physical memory. On macOS, Docker Desktop and Podman run containers inside a Linux VM, so container CPU and memory are relative to that VM. The collector retains up to 30 days in SQLite; the dashboard's own history is the samples captured while the tab is open.

Maturity

AreaStatus
Collector (Linux, macOS), API, history, TLS, proxyWorking; native runs on macOS 26 and Ubuntu 24.04, tested in CI
Docker and PodmanWorking; parsing tested from recorded output, plus real Docker and Podman checks in CI
DashboardWorking; Playwright on WebKit, Chromium and mobile WebKit
Installers (launchd, systemd) and remote deployWorking; dry-run checked in CI
Mac cluster (worker + coordinator, read-only)Experimental; tested with unit and mocked e2e tests and one real Mac enrolled over loopback. Not yet checked across two Macs or over HTTPS
Native Mac appBuilt and run on macOS 27.2 (Apple silicon) with live and demo data; alert logic and the window and sidebar flow have scripted tests (scripts/test-native-logic.sh, scripts/test-native-ui.sh), not run in CI. Not checked on macOS 26; the widget has not been seen in the widget gallery. the v0.7.0 DMG and pkg on the release are Developer ID signed and notarized (checked with spctl and stapler validate; the DMG and pkg carry the tickets, the app inside the DMG does not); builds from this repository's CI have no Apple secrets and are not; v0.3.0 DMG is Developer ID signed, not notarized (Apple silicon only)
Homebrew formulaStub, not published
Fleet Pro (agent/vytrix_fleet.py, cluster/fleet_coordinator.py)New and opt-in; unit-tested and covered by CI. Not yet checked across several Macs, over HTTPS, or with real webhook, SSO and Prometheus receivers
GPU, sensors, per-app networkNot collected by the base collector. Fleet Pro adds best-effort GPU and thermal data (Apple capability; Linux DRM where exported) and macOS per-process network via nettop; unavailable counters are omitted, and Linux per-process network is not provided

License

Vytrix is licensed under the Business Source License 1.1.

You may use Vytrix for development, evaluation, testing, demonstration, and other non-production purposes under the terms of the BSL.

Production and commercial deployments require a commercial license from Zyvor AI Labs Private Limited. The Supporter Commercial License is a one-time payment that grants it for your own machines: pay what you like, suggested ₹99 or US$1 (the page takes INR; a USD page is not set up yet). Companies, USD invoices and multi-entity licences: write to [email protected] and we will send an invoice. Using Vytrix without paying changes nothing in the app; nothing is locked.

Each version converts to the Apache License 2.0 four years after its initial release under the Business Source License.

See LICENSE for the complete terms, NOTICE and THIRD_PARTY.md. Releases up to and including v0.3.0 were published under Apache-2.0 and remain so.

Copyright © 2026 Zyvor AI Labs Private Limited · zyvor.dev

See what your machine is really doing. · Quickstart · Security · Changelog

Fleet Pro

Opt-in and still read-only (no command channel): richer silicon and AI workload telemetry, central history, tags and groups, auditor and editor roles, alerts and webhooks, an SSO-proxy endpoint, Prometheus export and local capacity forecasting. See docs/FLEET-PRO.md. The original collector and coordinator stay available as a rollback path.

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.