MXterm Pro

What MXterm writes

MXterm Pro edits a few files outside its own folder so that coding agents in its panes can reach its built-in server, and so that worktrees and Finder actions work. None of these files holds a live token: each pane's token reaches its agent only through that pane's environment. In Settings → Agents, Connect my agents to MXterm (on by default) stops every write to agent settings files, and each agent has its own Disconnect. Neither switch deletes anything beyond the agent's own mxterm entry, and the app removes nothing when you quit or uninstall. This page lists each file so you can check or remove it yourself.

Agent settings files

At each launch, while Connect my agents to MXterm is on, MXterm Pro adds one mxterm entry to the settings of Claude Code, Codex and omp (installed or not), and of the other agents below when they are installed and connected. The entry holds only placeholders or the local address of the built-in server, never a token.

FileWhat is added
~/.claude.json (or under $CLAUDE_CONFIG_DIR)mcpServers.mxterm with ${SWARMTERM_MCP_URL} and Bearer ${SWARMTERM_SESSION}, and crossSessionInbound: "accept" when that key is absent; created if missing
~/.codex/config.toml (or under $CODEX_HOME)[mcp_servers.mxterm] with only url = "http://127.0.0.1:<port>/mcp"; each Codex pane passes its own token when it starts
~/.omp/agent/mcp.jsonthe same mxterm entry as for Claude Code, with a timeout
Other agents, only when installedthe same single entry in ~/.gemini/config/mcp_config.json (Antigravity), ~/.config/opencode/opencode.json, ~/.qwen/settings.json, ~/.copilot/mcp-config.json, and, only if the file already exists, ~/.openclaw/openclaw.json, ~/.config/amp/settings.json, ~/.config/crush/crush.json and ~/.hermes/config.yaml; ~/.cursor/mcp.json only after you press Connect for Cursor
  • Backup: the first time an existing file is changed, a private copy is kept beside it as <file>.mxterm-backup.
  • Disconnect (per agent, in Settings → Agents) removes that agent's mxterm entry if it is still MXterm's own, and stops it coming back. It leaves the backup and crossSessionInbound in place.
  • Plugins (experimental, invited testers, macOS, off by default): turning a plugin on in Settings → Plugins adds its own entry, such as mxterm-hub, to the same files; turning it off removes it.
  • Older builds added a line to ~/.zshrc. 1.1.87 removes that line once, if it is still there, and writes nothing to ~/.zshrc.

In your projects

PathWhenWhat
<project>/.gemini/settings.jsonwhen a Gemini CLI pane starts there, with Connect my agents to MXterm onan mxterm entry with ${MXTERM_MCP_URL} and Bearer ${MXTERM_SESSION}, filled in by Gemini CLI from the pane's environment; no token is written. Builds before 1.1.29 wrote the token itself; the next Gemini pane replaces it.
<project>/.gitignorewith that Gemini pane, and whenever a worktree is created.gemini/settings.json and .worktrees/; the file is created if missing. The .worktrees/ line is added even with the switch off.
<project>/.worktrees/<branch>/when worktree isolation starts an agent pane, or an agent asks for a worktreea git worktree on its own branch
<project>/.mxterm/when an agent opens a pane with a brief (experimental, invited testers, Let agents open panes on; macOS and Linux)tasks/task-<id>.md, private to you (mode 0600), deleted when that pane closes; the folder keeps a .gitignore of its own containing *
<project>/.gitkeep and one commitonly in an existing repository with no commit yet, when a worktree is neededan "Initial commit" containing only .gitkeep, so a branch can be made

Claude Code extras

  • Status line: ~/.claude/settings.json gets a statusLine that runs MXterm Pro, on by default. Turn it off with Manage Claude Code's status line in Settings → General; that removes the entry. A status line you set yourself is never replaced.
  • Orchestrator skill: only when you click Install in Settings → Agents (experimental, invited testers, Let agents open panes on): six files and a manifest in ~/.claude/skills/mxterm-orchestrator/ and in ~/.agents/skills/mxterm-orchestrator/. An installed copy is refreshed at launch, never over a file you edited. The same place removes it.

On the system

  • Finder actions (macOS): ~/Library/Services/New MXterm Pro Tab Here.workflow and New MXterm Pro Window Here.workflow, checked at every launch and rewritten only if they changed. Delete them from that folder to remove them; they come back at the next launch.
  • Open at login: only when you turn it on; macOS uses the system login items, Windows the Run key, Linux an autostart entry.
  • Keychain: saved SSH passwords and keys, the licence, and a key for the agent ledger.
  • Temporary files: pasted images in $TMPDIR/mxterm-clipboard (removed after a day) and files opened over SFTP in $TMPDIR/mxterm-remote-files (pruned by age and size).

In each terminal pane's environment

Each terminal pane gets MXTERM_SESSION and SWARMTERM_SESSION (that pane's own token, random and kept only in memory) and MXTERM_MCP_URL and SWARMTERM_MCP_URL (the local address of the built-in server). The token ends with the pane. This is how the entries above work without holding a secret.

Inside the app's own folder

Everything else stays in MXterm Pro's own data folder: Control Room rooms and logs, agent pane settings, remote access and sharing state, the account and limits cache, the Token Meter cache, diagnostics logs, your switch choices, and the usage-report queue and switch. What the usage report sends is on the telemetry page.

KI zu MXterm Pro fragen

Öffnen Sie eine Live-Antwort in ChatGPT, Claude oder Perplexity, die auf dieser Seite basiert. Den Text schreibt der Assistent, nicht wir. Prüfen Sie daher die Quellen, die er nennt.

Gesendete FrageWas ist MXterm Pro, und für welche Art von Entwickler ist es gedacht?