Terminal Commands

Run commands on your machines

Terminal Commands connects your AI and compatible MCP clients to computers and servers you control. Choose a saved project, inspect files, edit bounded text files, run locally approved commands, and review command activity and source changes.

What you can ask

Show my connected machines.
Example path →

MCP connector

Server URL: https://mcp.terminalcommands.fr/mcp
Login: OAuth
Protocol: MCP 2026-07-28
Instructions: Routes requests only to machines paired with the authenticated user. Prefer dedicated read tools. Commands and writes are destructive and require explicit approval on the selected machine. Before other Terminal Commands tools for a new request, call once to show the inline Activity/Changes card. Reuse the conversationId and selected project returned by the previous request in this same conversation, even after it finished; only requestId changes. Continue the same request with its existing runId; and show its Activity at the current step in the chat. This applies to every Terminal Commands request, including device/status checks, folder reads and simple file writes. For general checks, use with {} or the active projectId; do not require project selection just to list devices. The active project stays selected for all later file and command requests in this conversation until the user explicitly changes it. Never choose a different project or fall back to the machine home folder or approved root. For an existing run, with runId first and reuse it. For a new file or coding request, FIRST with its original goal and one stable generated requestId, plus the projectId selected in app context/history or the exact projectName explicitly named by the user. That first call resolves the bookmark and captures the baseline; do not call or before opening the card. If awaitingProject is true or workspaceError is returned, ask the user to choose or recover the project in the card, then stop and wait. Do not execute commands or writes while waiting. Selection automatically resumes the same original request. On a project-selection continuation, call with the existing runId to show its Activity in the current reply. Do not create a new run or baseline for a continuation. Use for every command, including later commands in the same run; this displays the current Activity card. Submission briefly observes short commands. If still pending, running or cancelling, repeatedly call with the returned taskId, latest eventSequence as afterEventSequence and timeoutMs 15000. Keep this turn active until the command reaches completed, failed, cancelled or unknown, then handle the result and continue the remaining original goal. Results are returned server-side; a loaded Activity card or open chat page is not required. Never end the turn just to wait for a command. Closing the page is not a completion signal. If the host cancels or ends the turn, recover the existing task on the next authorized turn; the plugin cannot start a new ordinary Chat turn itself. Use dedicated file tools only inside the active project, and after edits. Keep request, run and operation IDs stable on retries. Never automatically replay an unknown command. A completed command does not prove the overall coding task is complete. When finished handling the original request, including file-only work or reporting a failure, call before your final reply. Never finish while a command is active or unconfirmed. For each new user request, open a fresh run with a new requestId, preserving the selected project and the conversationId returned by the first workspace request in this chat. Reuse that conversationId across project changes but never across different chats.

Tools

Reads data11

Changes data2

Deletes or can't be undone10

Skills

None in the store listings or on the server

MCP connector

Server URL

https://mcp.terminalcommands.fr/mcp

OAuth

Protocol

MCP 2026-07-28

Instructions

Routes requests only to machines paired with the authenticated user. Prefer dedicated read tools. Commands and writes are destructive and require explicit approval on the selected machine. Before other Terminal Commands tools for a new request, call once to show the inline Activity/Changes card. Reuse the conversationId and selected project returned by the previous request in this same conversation, even after it finished; only requestId changes. Continue the same request with its existing runId; and show its Activity at the current step in the chat. This applies to every Terminal Commands request, including device/status checks, folder reads and simple file writes. For general checks, use with {} or the active projectId; do not require project selection just to list devices. The active project stays selected for all later file and command requests in this conversation until the user explicitly changes it. Never choose a different project or fall back to the machine home folder or approved root. For an existing run, with runId first and reuse it. For a new file or coding request, FIRST with its original goal and one stable generated requestId, plus the projectId selected in app context/history or the exact projectName explicitly named by the user. That first call resolves the bookmark and captures the baseline; do not call or before opening the card. If awaitingProject is true or workspaceError is returned, ask the user to choose or recover the project in the card, then stop and wait. Do not execute commands or writes while waiting. Selection automatically resumes the same original request. On a project-selection continuation, call with the existing runId to show its Activity in the current reply. Do not create a new run or baseline for a continuation. Use for every command, including later commands in the same run; this displays the current Activity card. Submission briefly observes short commands. If still pending, running or cancelling, repeatedly call with the returned taskId, latest eventSequence as afterEventSequence and timeoutMs 15000. Keep this turn active until the command reaches completed, failed, cancelled or unknown, then handle the result and continue the remaining original goal. Results are returned server-side; a loaded Activity card or open chat page is not required. Never end the turn just to wait for a command. Closing the page is not a completion signal. If the host cancels or ends the turn, recover the existing task on the next authorized turn; the plugin cannot start a new ordinary Chat turn itself. Use dedicated file tools only inside the active project, and after edits. Keep request, run and operation IDs stable on retries. Never automatically replay an unknown command. A completed command does not prove the overall coding task is complete. When finished handling the original request, including file-only work or reporting a failure, call before your final reply. Never finish while a command is active or unconfirmed. For each new user request, open a fresh run with a new requestId, preserving the selected project and the conversationId returned by the first workspace request in this chat. Reuse that conversationId across project changes but never across different chats.

Tools