Skip to content
ReviewRamble

The weekly brief

AI, reviewed, in your inbox.

  • New reviews of tools, models, MCP servers and repos, with the verdict up front
  • Re-checked prices and ratings when something changes
  • New analysis pieces as they are published

Models, agents, MCP servers and the tools actually worth your time.

One email a week. Unsubscribe with the link in any brief. How we use your email

uniTerm

Free all-in-one client for SSH, RDP, SFTP and databases, with a bring-your-own-key AI agent and an MCP bridge.

Reviewed by ReviewRamble editors · Updated

View on GitHub Open source

Share this page:XLinkedInFacebookRedditWhatsAppEmail

Our rating

3.8/5

uniTerm's MCP bridge is the most carefully built thing in it; the in-terminal agent is useful but runs ordinary writes without asking unless you tighten the default.

What is uniTerm?

uniTerm is a Go and Vue desktop app, built on Wails v3, that puts more than 30 protocols in one window: SSH, Telnet, Mosh and serial terminals; SFTP, SCP, FTP, SMB, WebDAV and S3 file transfer; RDP, VNC, SPICE and X11 remote desktops; MySQL, PostgreSQL, Oracle, SQL Server, SQLite, Redis, MongoDB and Elasticsearch clients; and Kubernetes, Docker, Podman and nerdctl. It runs on Windows, macOS, Linux and Android, and as a multi-tenant Docker server you open in a browser.

The AI side has three parts. A sidebar agent plans and runs shell commands in your active terminal over several rounds, using any Anthropic- or OpenAI-compatible API you configure. The database client turns plain-language questions into queries. And an optional built-in MCP server lets outside agents such as Claude Code, Codex or Gemini CLI run commands and move files on your saved connections without ever seeing the passwords.

It is a personal hobby project under Apache 2.0, created in May 2026, with frequent releases (v1.12.3 on 15 November 2026). The author says there are no plans to commercialise it.

Is uniTerm worth it?

uniTerm's best AI idea is the reverse of what most terminals do. Instead of bolting a chatbot on, it lets coding agents you already use reach your servers through it. Its MCP server, off by default, listens on 127.0.0.1:61207, needs a per-client bearer token stored only as a SHA-256 hash, asks you to approve every command by default, auto-denies after 110 seconds, caps runs at eight concurrent commands with a five-minute timeout, and logs each call to a JSONL audit file. The agent sees connection names, never the passwords or keys.

The sidebar agent is a different trust model. It sends your request and the terminal output it reads to whichever model endpoint you configure, then runs commands in your live tab for up to 20 rounds by default. Each command's risk is the higher of what the model declares and a local pattern check, which the code itself calls defence in depth rather than a wall. The default mode, Confirm Dangerous, only stops for high-risk commands, so anything graded a write, such as moving files or installing packages, runs without a prompt. Switch to Confirm Write or Confirm All before pointing it at production.

Credential storage is stated plainly in the code. Passwords, private keys, TOTP secrets, proxy passwords and AI API keys sit in local JSON files with each secret field encrypted with AES-GCM. The key is either a random key kept in the OS keychain or one derived with PBKDF2 from a master password you set. Cloud sync to Git, WebDAV, SFTP, S3 or SMB encrypts on the client with AES-256-GCM before upload.

The catch: this is one maintainer's hobby project, now with 1,217 stars, 85 open issues and a README that welcomes vibe-coded contributions. Windows builds are not code-signed, so antivirus tools may flag them. Turning on remote MCP access switches the listener to 0.0.0.0 over plain HTTP, so tokens cross the network unencrypted unless you tunnel through SSH as the docs suggest.

Our take: as a free stand-in for all-in-one clients like MobaXterm or Xshell, with an AI agent bolted on, it is generous; the part worth adopting today is the MCP bridge, because it keeps your agents away from your keys.

Where it earns its place

  • A DevOps engineer at a 30-person SaaS company using Claude Code

    Let the agent check nginx config and logs on staging servers without giving it SSH keys

    Enable uniTerm MCP, generate a token named claude-code, and add the loopback URL to Claude Code. Each command shows in an approval dialog with the full text, and the audit log records the outcome. Costs only your existing Claude plan.

  • A Windows sysadmin at a small managed-services firm

    Replace several separate tools for SSH, RDP, SFTP and SQL Server

    Import saved sessions from MobaXterm or Xshell, then use one window with split panes for terminals, desktops and database tabs. Free; the AI sidebar is optional and needs your own API key.

  • A backend developer debugging a PostgreSQL replica

    Ask the database client questions in plain language instead of writing SQL

    Configure a model in Settings and use the natural-language query feature in the database tab. The request goes to the model provider you chose, so a local or company gateway suits sensitive data.

Before you commit

Setup
Install from GitHub Releases, Scoop, Homebrew or a .deb/.rpm; add an AI model with endpoint, model name and key in Settings, then open the AI sidebar in any terminal tab.
Integrations
Anthropic Messages, OpenAI Chat Completions and Responses APIs; built-in MCP server over Streamable HTTP for Claude Code, Codex, Gemini CLI and others; connects to external stdio, HTTP or SSE MCP servers as tools for its own agent.
Data and privacy
Connections and secrets stay local, with secret fields encrypted by AES-GCM under a keychain-held or PBKDF2-derived key. AI requests carry your prompts and terminal output to the provider you configure. Sync is client-side encrypted with AES-256-GCM. The Docker server keeps master keys only in memory.
Cost at scale
Free. AI cost is your provider's token bill, which grows with agent rounds (default limit 20, configurable, 0 for unlimited).
Limits
Agent risk grading is pattern matching plus the model's own declaration; MCP exec has a concurrency cap of 8 and a five-minute timeout per command; read_remote_file reads at most 256KB; MCP is disabled in the Docker server build.

What’s good

  • SSH, RDP, SFTP, databases and Kubernetes in one free app
  • MCP server keeps credentials inside the app, with approvals and an audit log
  • Secrets encrypted at rest with AES-GCM; sync encrypted on the client
  • Any Anthropic- or OpenAI-compatible model, including local gateways

What’s not

  • Agent's default mode runs commands graded as writes without asking
  • Terminal output goes to whichever model provider you configure
  • One-maintainer hobby project with 85 open issues
  • Windows builds are unsigned; remote MCP is plain HTTP

Best for

Sysadmins and DevOps engineers who juggle SSH, RDP and database sessions and want an AI agent or Claude Code to work on those hosts without handling the credentials.

uniTerm alternatives

uniTerm: common questions

Is uniTerm worth it?
uniTerm's MCP bridge is the most carefully built thing in it; the in-terminal agent is useful but runs ordinary writes without asking unless you tighten the default. We rate it 3.8 out of 5.
How much does uniTerm cost?
Free under Apache 2.0. AI features use your own API key, billed by the provider you choose.
What are the downsides of uniTerm?
Agent's default mode runs commands graded as writes without asking; terminal output goes to whichever model provider you configure; one-maintainer hobby project with 85 open issues; windows builds are unsigned; remote MCP is plain HTTP.
Who is uniTerm best for?
Sysadmins and DevOps engineers who juggle SSH, RDP and database sessions and want an AI agent or Claude Code to work on those hosts without handling the credentials.
How much does uniTerm cost for a team or in production?
Free. AI cost is your provider's token bill, which grows with agent rounds (default limit 20, configurable, 0 for unlimited).
Is uniTerm safe for company data?
Connections and secrets stay local, with secret fields encrypted by AES-GCM under a keychain-held or PBKDF2-derived key. AI requests carry your prompts and terminal output to the provider you configure. Sync is client-side encrypted with AES-256-GCM. The Docker server keeps master keys only in memory.
What are the best uniTerm alternatives?
The closest alternatives we have reviewed are Stakpak, herdr and OpenCode.

Answers checked Nov 18, 2026.