Self-Hosted Package Installation Guide

Overview

The ManageLM installer is a single self-extracting .sh.gz file that includes the portal, agent source, and a bundled Node.js runtime. No prerequisites need to be installed on the target server — just download, extract, and run.

The installer handles both first install and upgrades. On upgrade, your configuration (.env) is preserved automatically.

Requirements

• Linux (x64 or arm64) — Debian, Ubuntu, RHEL, Rocky, AlmaLinux, etc.
• PostgreSQL 15+ — application database
• Redis 7+ (or Valkey) — sessions, pub/sub, rate limiting
• 1 GB RAM minimum
• Two HTTPS endpoints:
  • Portal URL (SERVER_URL, e.g. https://managelm.example.com) — admin UI + agent WebSocket. Can be internal-only. The certificate just needs to be trusted by every agent host: a public CA (Let's Encrypt etc.) works without setup; a private/internal CA works too if you distribute its root cert to the agents' system trust store before enrolling them. See Internal Portal URL.
  • MCP public URL (PUBLIC_URL, e.g. https://mcp.example.com) — required for Claude.ai Custom Connector. Must be a dedicated subdomain on a public DNS record, reachable from the internet, with a certificate from a public CA (Anthropic's servers won't accept a private CA). See MCP Public URL.

Node.js is bundled in the installer — no system packages required beyond the database and cache.

Download

Download the latest release for your architecture from GitHub Releases.

Then extract and run:

gunzip managelm-<version>-linux-<arch>.sh.gz

Install

Run the installer as root:

sudo bash managelm-<version>-linux-<arch>.sh

The installer will prompt for:

For unattended installation, use --yes to accept all defaults:

sudo bash managelm-<version>-linux-<arch>.sh --yes

Other flags:

  --install-dir DIR   Override install directory (default: /opt/managelm-server)
  --user USER         Service user (default: managelm)
  --port PORT         Portal port (default: 3000)
  --no-systemd        Skip systemd service registration

Configure

Set up PostgreSQL and Redis first — see PostgreSQL Setup and Redis Setup below for the full procedure (creating the user, granting schema access, enabling persistence). The portal will fail to start until both are reachable with valid credentials.

Once PostgreSQL and Redis are running, edit the .env file:

sudo vi /opt/managelm-server/portal/.env

At minimum, set these values:

# How browsers and agents reach the portal
SERVER_URL=https://managelm.example.com

# Public MCP URL for Claude.ai (dedicated subdomain, public CA)
# Required for the Claude Custom Connector flow; leave unset for testing.
# See "MCP Public URL" below for the reverse-proxy setup.
PUBLIC_URL=https://mcp.example.com

# PostgreSQL connection (created in the PostgreSQL Setup section below)
DATABASE_URL=postgresql://managelm:your-password@localhost:5432/managelm

# Redis connection
REDIS_URL=redis://localhost:6379

# Sender email (only effective when SMTP_HOST is also set)
[email protected]

Then start the service:

systemctl start managelm-server

Check it's running:

systemctl status managelm-server
curl -s http://localhost:3000/health

First Steps After Install

Once the portal is running:

1. Front the portal with TLS — reverse proxy at SERVER_URL (nginx or Apache). Required before agents can enroll.
2. Set up the MCP public URL — dedicated subdomain (e.g. mcp.example.com) reverse-proxied to portal /public/. Required for Claude.ai Custom Connector.
3. Register your account at your SERVER_URL. The first user becomes the account owner.
4. Configure the LLM — Local (Ollama), Cloud, or Proxied access mode.
5. Import skills from the built-in catalog.
6. Install an agent on your first server.
7. Connect Claude via MCP.

Internal Portal URL

Place a reverse proxy in front of the portal for TLS. This fronts SERVER_URL — admin UI, agent WebSocket, OAuth login/consent. The separate public-facing MCP endpoint is covered in MCP Public URL below.

WebSocket support is required for agent connections.

nginx

server {
    listen 443 ssl http2;
    server_name managelm.example.com;

    ssl_certificate     /etc/ssl/certs/managelm.pem;
    ssl_certificate_key /etc/ssl/private/managelm.key;

    location / {
        proxy_pass http://127.0.0.1:3000;
        proxy_http_version 1.1;
        proxy_set_header Upgrade    $http_upgrade;
        proxy_set_header Connection "upgrade";
        proxy_set_header Host              $host;
        proxy_set_header X-Real-IP         $remote_addr;
        proxy_set_header X-Forwarded-For   $proxy_add_x_forwarded_for;
        proxy_set_header X-Forwarded-Proto $scheme;
        proxy_read_timeout 86400s;
        proxy_send_timeout 86400s;
    }
}

Apache

<VirtualHost *:443>
    ServerName managelm.example.com

    SSLEngine on
    SSLCertificateFile    /etc/ssl/certs/managelm.pem
    SSLCertificateKeyFile /etc/ssl/private/managelm.key

    ProxyPreserveHost On
    ProxyPass        / http://127.0.0.1:3000/
    ProxyPassReverse / http://127.0.0.1:3000/

    RewriteEngine On
    RewriteCond %{HTTP:Upgrade} =websocket [NC]
    RewriteRule /(.*) ws://127.0.0.1:3000/$1 [P,L]

    ProxyTimeout 86400
</VirtualHost>

MCP Public URL

Claude.ai's Custom Connector flow runs OAuth and every MCP tool call from Anthropic's backend, so the MCP endpoint has to be reachable from the public internet. ManageLM exposes a small curated /public/ surface on the portal — OAuth discovery, token exchange, MCP, the connector logo, and (when you opt-in per-agent) the agent bootstrap + WebSocket so agents outside your network can enroll. Everything else (admin UI, agent UI, login, internal APIs) stays on SERVER_URL and is not reachable from the public hostname.

You point Claude (and any external agents) at a dedicated public subdomain that reverse-proxies to http://127.0.0.1:3000/public/. Single rule, allowlist enforced by the portal.

Requirements

• Dedicated subdomain — e.g. mcp.example.com. Don't share with the portal or other services; the well-known OAuth discovery URL must sit at the root of this hostname.
• Public DNS — the name must resolve from outside your network (Anthropic's backend fetches it).
• Trusted public CA — Let's Encrypt or commercial. Private CAs are rejected by Anthropic's servers.
• Port 443 open from the internet to your reverse proxy.

Set PUBLIC_URL

In /opt/managelm-server/portal/.env:

PUBLIC_URL=https://mcp.example.com

Then restart: systemctl restart managelm-server. The portal advertises this URL in OAuth metadata and shows it in Settings → Claude Connectors as the URL users paste into Claude.

nginx

server {
    listen 443 ssl http2;
    server_name mcp.example.com;

    ssl_certificate     /etc/letsencrypt/live/mcp.example.com/fullchain.pem;
    ssl_certificate_key /etc/letsencrypt/live/mcp.example.com/privkey.pem;

    location / {
        proxy_pass http://127.0.0.1:3000/public/;
        proxy_set_header Host              $host;
        proxy_set_header X-Real-IP         $remote_addr;
        proxy_set_header X-Forwarded-For   $proxy_add_x_forwarded_for;
        proxy_set_header X-Forwarded-Proto $scheme;
        proxy_set_header X-Forwarded-Host  $host;
    }
}

Apache

<VirtualHost *:443>
    ServerName mcp.example.com

    SSLEngine on
    SSLCertificateFile    /etc/letsencrypt/live/mcp.example.com/fullchain.pem
    SSLCertificateKeyFile /etc/letsencrypt/live/mcp.example.com/privkey.pem

    ProxyPreserveHost On
    ProxyPass        /  http://127.0.0.1:3000/public/
    ProxyPassReverse /  http://127.0.0.1:3000/public/
</VirtualHost>

Verify

# OAuth discovery — must return JSON metadata
curl -s https://mcp.example.com/.well-known/oauth-authorization-server | head

# MCP endpoint — must return 401 with WWW-Authenticate
curl -i -X POST https://mcp.example.com/mcp -H 'Content-Type: application/json' -d '{}'

# Anything outside the allowlist — must return 404
curl -i https://mcp.example.com/api/auth/me

External agents

The same /public/ surface also serves the agent bootstrap and WebSocket, so agents that can't reach SERVER_URL (remote hosts, customer servers, anything outside your VPN) can enroll and connect via the public hostname instead.

In Agents → Add Agent (or Agent detail → Reinstall), flip the Use Public URL switch before copying the install command. The toggle is only shown when PUBLIC_URL is configured. With it on, the generated curl/wget/iex command points at mcp.example.com and the agent's config.yaml stores that URL as its server_url — the agent uses it for bundle download, enrollment polling, updates, and the runtime WebSocket. With it off, the install uses SERVER_URL (the original behaviour).

Threat-alert actions

When PUBLIC_URL is set, the one-click buttons in threat-alert emails (Kill Service / Kill Session / Discard Alert) are served via the public hostname too, so an admin can act on an alert from anywhere — no VPN needed. The authorization is the single-use signed token in the link, not a portal session, and the confirmation page is fully self-contained. No extra proxy rule is required — the existing / → /public/ rule already covers it. The “Open the alert in the portal” link in the same email still points at SERVER_URL, since it loads the admin UI and stays internal.

The choice is per-install: re-enrolling an existing agent re-prompts, so an agent that was originally internal can move to public (or vice versa) without portal-side surgery.

SMTP / Email

The portal sends emails for account verification, password resets, and team invitations. When SMTP_HOST is not set, email sending is disabled entirely — no connection attempts are made.

To enable emails, configure an SMTP relay (Brevo, Mailgun, SendGrid, etc.):

SMTP_HOST=smtp.brevo.com
SMTP_PORT=587
SMTP_SECURE=starttls
SMTP_USER=your-username
SMTP_PASS=your-password

PostgreSQL Setup

ManageLM requires PostgreSQL 15+ (16 recommended). The portal runs migrations automatically on startup and needs full control over its schema.

1. Create the database and user

   # Connect as a PostgreSQL superuser
   sudo -u postgres psql
   
   -- Create the ManageLM user
   CREATE USER managelm WITH PASSWORD 'your-strong-password';
   
   -- Create the database owned by that user
   CREATE DATABASE managelm OWNER managelm;
   
   -- Connect to the new database
   \c managelm
   
   -- Grant full schema permissions (required for migrations)
   GRANT ALL ON SCHEMA public TO managelm;
   
   -- PostgreSQL 15+ changed defaults — grant CREATE explicitly
   ALTER DEFAULT PRIVILEGES IN SCHEMA public
     GRANT ALL ON TABLES TO managelm;
   ALTER DEFAULT PRIVILEGES IN SCHEMA public
     GRANT ALL ON SEQUENCES TO managelm;
   
   \q

2. Allow remote connections (if PostgreSQL is on a different host)

   # In postgresql.conf
   listen_addresses = '*'
   
   # In pg_hba.conf — allow the portal host
   host  managelm  managelm  10.0.0.0/8  scram-sha-256

   Restart PostgreSQL: systemctl restart postgresql

3. Set DATABASE_URL in .env

   DATABASE_URL=postgresql://managelm:your-strong-password@localhost:5432/managelm

4. Test the connection

   psql "postgresql://managelm:your-strong-password@localhost:5432/managelm" -c "SELECT 1;"

Redis Setup

ManageLM requires Redis 7+ (or Valkey). It is used for real-time agent communication, session state, and pub/sub — it must be available at all times.

1. Enable persistence (recommended)

   # In redis.conf
   appendonly yes
   appendfsync everysec

2. Set REDIS_URL in .env

   # Without authentication
   REDIS_URL=redis://localhost:6379
   
   # With authentication (Redis 6+ ACL)
   REDIS_URL=redis://username:password@redis-host:6379

3. Test the connection

   redis-cli ping
   # Expected: PONG

Environment Variables

All settings are in /opt/managelm-server/portal/.env. The full reference is the same as the Docker guide.

Required

Strongly recommended

Set this before adding cloud connectors, configuring an LLM API key, enrolling agents, or generating a PKI CA — the portal starts without it (warning only), but any feature that encrypts data at rest will fail until it's configured. Rotating the key invalidates existing encrypted rows, so set it once and keep it.

Database & Redis

SMTP & DKIM

Optional

Advanced

File Layout

/opt/managelm-server/
├── nodejs/          # Bundled Node.js runtime
│   └── bin/node
├── portal/          # ManageLM portal application
│   ├── dist/        # Compiled backend (bytecode)
│   ├── web/dist/    # Frontend build
│   ├── node_modules/
│   ├── cluster.cjs  # Multi-worker entry point
│   ├── loader.cjs   # Bytecode loader
│   ├── .env         # Configuration (root:managelm 640)
│   └── ...
└── agent/           # Agent source (served to managed hosts)
    ├── managelm-agent.py
    ├── lib/
    └── bin/

All files are owned by root:root except .env which is root:<service-user> 640. The service user can read the configuration but cannot modify any binaries.

Upgrading

Download the new release and run it on the same server. The installer detects the existing installation and upgrades in place:

gunzip managelm-<version>-linux-<arch>.sh.gz
sudo bash managelm-<version>-linux-<arch>.sh

The upgrade will:

• Stop the running service
• Replace compiled code, static assets, and the bundled Node.js runtime
• Reinstall production dependencies
• Update the systemd unit file
• Preserve your .env (never overwritten)
• Restart the service

Database migrations run automatically on startup.

Backup & Restore

PostgreSQL holds the durable data and is the only thing you strictly need to back up. Redis is rebuilt from PostgreSQL on portal startup; back it up only if you want to preserve in-flight sessions across restores. Don't forget to copy .env — it holds ENCRYPTION_KEY and is not regenerable.

Backup

# PostgreSQL (run regularly)
pg_dump -U managelm managelm > managelm-backup-$(date +%F).sql

# Configuration (contains ENCRYPTION_KEY — store securely)
cp /opt/managelm-server/portal/.env ./env-backup-$(date +%F)

# Optional — Redis snapshot (in-flight session state)
redis-cli BGSAVE
cp /var/lib/redis/dump.rdb ./redis-backup-$(date +%F).rdb

Restore

# Stop the portal
systemctl stop managelm-server

# Restore PostgreSQL
psql -U managelm managelm < managelm-backup-YYYY-MM-DD.sql

# Restore configuration (if needed)
cp env-backup-YYYY-MM-DD /opt/managelm-server/portal/.env

# Start the portal — Redis state rebuilds automatically (users will need to log in again)
systemctl start managelm-server

Monitoring

# Service status
systemctl status managelm-server

# Health check
curl -s http://localhost:3000/health

# Live logs
journalctl -u managelm-server -f

# Last 100 lines
journalctl -u managelm-server -n 100

Uninstall

# Stop and disable the service
systemctl stop managelm-server
systemctl disable managelm-server
rm /etc/systemd/system/managelm-server.service
systemctl daemon-reload

# Remove the install directory
rm -rf /opt/managelm-server

# (Optional) Remove the service user
userdel managelm

This does not remove PostgreSQL or Redis data. Drop the database manually if no longer needed:

sudo -u postgres dropdb managelm
sudo -u postgres dropuser managelm

Troubleshooting

Portal won't start

Agents can't connect

Database permission errors

Email not sending

.env permission denied