Skip to content

Editors and chat apps ​

orch works from any app that supports MCP (Model Context Protocol), not only from the command line: VS Code with Copilot (agent mode), Cursor, and Claude Desktop. The app starts orch-mcp itself and gets the orch_run and orch_status tools plus three ready-made prompts. You do not run orch by hand.

Install orch first (npm install -g claude-codex-orch), then add it to your app.

One command ​

bash
orch mcp-config vscode            # print the JSON and the file it belongs in
orch mcp-config vscode --write    # add the entry to that file

Use cursor or claude-desktop instead of vscode for those apps. --write keeps every other server in the file, says unchanged when orch is already there, and refuses (changes nothing) when the file has comments or is not plain JSON; then paste the printed entry in by hand. Restart the app afterwards.

AppFileScope
VS Code (Copilot agent mode).vscode/mcp.json in the project (--cwd DIR picks the project)one project
Cursor~/.cursor/mcp.jsonevery project
Claude DesktopmacOS ~/Library/Application Support/Claude/claude_desktop_config.json, Windows %APPDATA%\Claude\claude_desktop_config.json, Linux ~/.config/Claude/claude_desktop_config.jsonevery chat

By hand, the VS Code entry is:

json
{ "servers": { "orch": { "type": "stdio", "command": "orch-mcp", "args": [] } } }

Cursor and Claude Desktop use mcpServers instead of servers and no type:

json
{ "mcpServers": { "orch": { "command": "orch-mcp", "args": [] } } }

Using it ​

  • Ask in the chat: "use orch to fix the typo in README.md". The assistant calls orch_run; orch picks Claude Code or Codex, fails over on a rate limit, and returns the answer with the agent, tokens and any roles used.
  • The prompts orch-fix, orch-review and orch-explain appear in the app's prompt menu (in VS Code, type / in chat and look for mcp.orch). Each takes one task and tells the assistant exactly how to call orch_run: orch-fix uses the developer role, orch-review the reviewer role, and orch-explain asks one question on Claude without editing.
  • Any option from Usage (role, team, auto, only, context, dryRun) can be requested in plain words or passed as tool arguments.

Check it works ​

  1. Run orch mcp-config <app> and make sure the file it names contains the entry.
  2. Restart the app, then open its list of tools or MCP servers: orch should show as running with orch_run and orch_status.
  3. Ask for a dry run: "call orch_run with dryRun true, roles developer and tester, and show the plan". It uses no tokens.

If the app cannot find orch-mcp, it was installed in a place the app's PATH does not include (common with nvm). Put the full path in command: which orch-mcp shows it.

What does not work ​

Web and mobile chat (claude.ai in a browser, phone apps) cannot start programs on your computer, so they cannot use orch. That needs a remote HTTP server with sign-in, which orch does not provide.