Termexo 0.10.10: let coding agents query the workbench through local MCP

I maintain Termexo, an MIT-licensed Windows workbench for coding agents. Version 0.10.10 adds 19 local MCP tools for terminals, tasks and selected settings.

New Claude Code, Codex, OpenCode, Grok Build and Antigravity terminals launched in Termexo can automatically load the connection after you enable and apply MCP settings. Existing agent processes need to restart, and the desktop must stay open.

A practical first trial: ask termexo-desktop to list workspaces, with no terminal input or settings changes. Then read a test terminal you choose and create a draft task with acceptance criteria you can inspect.

The server starts disabled and listens only on loopback. Terminal, task and settings permissions are separate; settings access needs enabling. The stdio bridge reads the current credential for each request rather than putting the token in generated configuration.

Automatic configuration has different scope: Claude Code, Codex and OpenCode use launch configuration; Grok Build and Antigravity use user-level configuration that other new clients may also see. Termexo removes its own additions when disabled and preserves other services.

Integration evidence: official MCP SDK handshake and discovery of 19 tools; an automatically connected Codex task terminal made a real workspace_list call. Other clients have different validation depths. The Codex trial required an extra carriage return; fully automatic task prompt submission still needs verification. Antigravity does not execute task-board tasks. A successful tool response is not acceptance of the resulting code.

Which workbench operation would you delegate first: reading terminal output, creating a draft task, or changing a selected setting?

Release and Windows installers · Setup and test boundaries

Maintainer post prepared with AI assistance. The diagram illustrates the implemented connection, not product UI.