Managing other AI Agents
DMJBot can run other AI coding agents for you and coordinate their work. Tools like Claude Code, GitHub Copilot CLI and aider are command-line agents that take a task in plain language and work on your code or files on their own. DMJBot can hand a task to one of them, let it work, and bring back the result — so DMJBot acts as the manager and the CLI agents are the workers.
This is powered by the Agent MCP server: you install it on a device next to an agent CLI and tell it how to launch that agent. From then on DMJBot can delegate tasks to it. See the Agent MCP Server page for the full setup and configuration.
Why this is useful
- Delegate big jobs. Instead of doing a large coding change inside the chat, DMJBot can pass it to a specialized coding agent that runs in your project and edits files directly.
- Use many agents on many machines. Connect an agent on your Mac, another on a Linux server, another on a work laptop — each in its own project folder. DMJBot can drive them all from one conversation.
- Run work in parallel. Several agents can work at the same time on different machines while you keep chatting.
- Mix agents and tools. Combine the coding agents with your other tools — email, file storage, Slack, Jira — in a single workflow.
How it works
- On each device, install the Agent MCP server through the
DMJBot Bridge and point it at the agent CLI (for example
claude -p 'PROMPT' --dangerously-skip-permissions) and a workspace folder. - Give each instance a clear title — e.g. "Claude on Mac", "Copilot on Linux server" — so DMJBot knows which one you mean.
- In chat, ask DMJBot to assign work to a specific agent. Long runs happen in the background, and you can follow progress while they work.
Because these agents run unattended, they are configured to act without asking for confirmation — see Agent MCP Server for what that means and how to keep it safe.
Example
With Claude Code set up on your Mac and GitHub Copilot CLI on a Linux server, you can orchestrate both in one message:
"Claude on my Mac should implement the waiting-list feature on the
signupbranch, and Copilot on the Linux server should finish the database migration script. Tell me when each one is done and summarize what it changed."
DMJBot dispatches each task to the right agent on the right machine, lets them work in parallel, and reports back as each finishes.
See also
- Agent MCP Server — full configuration, prompt template, progress and cancellation.
- Devices — connecting the machines your agents run on.
- DMJBot Bridge — installing the agent server on a device.
- Tools and MCP