Field noteNov 16, 20265 min read1 source

Mercury Agent 1.5.0 launches Mercury Desktop for macOS, Windows and Linux

In one minute

Mercury Agent 1.5.0 introduces Mercury Desktop for macOS, Windows and Linux, with options to run locally or connect to servers. Its release also adds bot management, approval controls, recoverable updates and MCP support, making the desktop app a new operating surface for teams running Mercury agents.

What changed: Mercury Agent 1.5.0 launches a cross-platform Mercury Desktop app for macOS, Windows, and Linux.

What happened

Mercury Agent’s v1.5.0 release, published on November 3, 2026, introduces Mercury Desktop for macOS, Windows and Linux. The app can install and start Mercury on the local computer or connect to multiple Mercury servers, with server switching from a top bar. The release describes the desktop as a unified place for chat, code, bots and research. It is available at mercuryagent.sh/download. The first release is not code-signed: macOS requires a one-time approval in System Settings, while Windows SmartScreen requires the user to continue through its warning. The release also gives a shell command for installing on Mac.

The desktop brings a bot inbox with run reports and tools to create or edit bots in natural language. Users can answer a bot’s question during a run with “Allow once,” “Always allow” or “Deny.” They can design a fleet’s crew or let Mercury choose it. Commands and skills can be inserted from the composer using / and #; saved chats, a microphone and Optimus permissions are also available across operating systems.

The release describes separate update flows for the desktop and servers. A desktop update downloads in the background, resumes after a dropped connection, is verified before installation and installs on restart or quit. For servers, mercury upgrade and the desktop updater wait for running work, download with checksums, resume interrupted downloads, back up, install and check the new version. If the update fails its check, the prior version and settings are restored automatically. Users can check the app and every server in Settings → About and choose “Update everything.” Required updates, such as a security fix, show a full-screen notice; other updates can be dismissed.

For remote access, a new phone or laptop receives a code that must be approved in Mercury Desktop, through /server, or in Telegram or WhatsApp. Sensitive actions—including shell access, plugins, channel tokens, restarts and server updates—require approval from a remote device; remote shell is off by default. The release says unused devices expire after 30 days, revoking a device cuts it off immediately, and a known device connecting from a new address triggers a notice. Mercury also adds MCP server support through mercury mcp add, chat attachments for photos and PDFs, and a research-run interface with an approvable plan, live progress, failure reasons and report export to PDF or Word.

What the archive adds

New Runtime covered Mercury Agent 1.4.0’s plugin system on October 24, 2026, in an earlier Field Note. Version 1.5.0 is a new release event, not a restatement: it adds a desktop operating surface and describes update, remote-access and bot controls beyond that earlier plugin release.

Trend signal

This release supports agent security moves to runtime boundaries: Mercury’s device approvals, default-off remote shell and confirmation for sensitive remote actions place control in the operating environment rather than relying only on prompts. The release describes safeguards, but does not establish their effectiveness in production.

Why now

The desktop release packages local use and multi-server access alongside bot workflows and update controls. The timing is a product change rather than a response to a stated external event; the release presents the app as Mercury’s new desktop interface.

Why it matters

For teams operating agents, the interface around a model is part of the control plane. Mercury’s local-or-server choice offers one way to keep desktop interaction while connecting to remote deployments; its device approval and remote-shell defaults make access decisions visible rather than implicit. Automatic rollback and update verification address a practical operational risk: an agent service can be difficult to recover if an upgrade fails while work is in progress.

These are useful design choices to evaluate, not proof that a deployment is secure or reliable. Teams should map them against their own identity, approval and recovery requirements. In particular, test whether users can distinguish safe routine operations from privileged actions, and confirm that rollback preserves the data and configuration the team needs. MCP support also widens the tool surface, increasing the importance of explicitly managing which servers are connected.

Limits and uncertainty

The release notes describe intended behavior but do not report independent testing, adoption, reliability measurements or security-audit results. They do not quantify how long server updates take, how often automatic rollback succeeds, or what protections apply to data transferred between a desktop and a server. The first desktop release is not code-signed, and its notes describe the user steps needed to bypass operating-system warnings. Teams should verify installation provenance, inspect the actual permission prompts and test recovery in a non-production environment before relying on these controls.

Skill to build now

This week, write a short acceptance test for an agent desktop or remote runner: enroll a new device and confirm it cannot connect before approval; attempt a privileged remote action and verify the approval prompt; then simulate a failed update on a disposable server and confirm the old version and settings return. Mercury’s release names these mechanisms, so they can be checked directly rather than inferred from a successful demo.

Reader takeaway

Mercury Agent 1.5.0 is more than a desktop wrapper: it combines local or server-backed operation with bot management, guarded remote access and update recovery. Builders considering it should test the access and rollback behavior themselves before treating the release’s safeguards as production guarantees.

Evidence / sources

  1. 1github.com/cosmicstack-labs/mercury-agent/releases
All Field Notes

Next reads

Read next

Open the graph
  1. 01topicAgentsExplore the agents topic hub.
  2. 02topicAgent InterfacesExplore the agent-interfaces topic hub.
  3. 03archiveField NotesOpen the latest editorial analysis.
  4. 04sourcesSourcesSee the original sources this archive cites.
Who read this page?Machine requests, hidden until opened

Loading the privacy-safe route aggregate...

Open the JSON contract