create_agentLABSCO SUMMARY
Isolation is by process and by path, not by convention.
Running several agents at once usually means several terminal windows and a lot of remembering which is which. Binding a session to a root path means an agent physically cannot wander into another project's code, and each one being its own process means one going wrong does not take the others with it. It is macOS and iTerm2 only — that is not a configuration detail, it is the architecture.
What it is
An orchestration server for coding agents on macOS. Each agent gets its own iTerm2 tab and its own process; sessions are bound to a root path so an agent cannot reach another session's code. Agent state persists across iTerm restarts.
What you get
- create_agent and delete_agent for adding and removing agent instances
- create_session tied to a root file path, with delete_session removing the session and every agent in it
- get_session_status for what every agent in a session is doing
- send_message_to_agent, which prepends the shared workflow prompt before delivering
- start_new_agent_conversation and clear_agent_conversation — a fresh tab, or closing the current one
- Process-level isolation between agents, with session boundaries stopping cross-codebase access
- Persistent state that survives an iTerm restart, and audit logging of agent interactions
- Limits on how many agents and sessions can exist, set as launch arguments
Requirements
macOS with iTerm2, since tabs are how agents are hosted. Python installed with uv, and registration by absolute path with the source directory on the Python path. The maximum agent and session counts, and the security level, are set as arguments when the server is launched.
Setup effort
Build from source
USE CASES
01
Running two pieces of work at once without them colliding
Create a session per codebase and put each agent in its own, bound to that root path.
02
Checking on everything from one place
Read the status of every agent in the session instead of clicking through tabs.
03
Giving an agent its next instruction
Send the message and let the shared workflow prompt be prepended for you.
04
Starting an agent over with a clean context
Close its current tab and open a new conversation for it.
05
Tearing down when the work is finished
Delete the session and every agent under it in one call.
8 tools
Listed in project docs. Your client may show a different list.
delete_agentcreate_sessionget_session_statusdelete_sessionsend_message_to_agentclear_agent_conversationstart_new_agent_conversationBEFORE YOU INSTALL
What to know first
Tools you get
8 tools named in the project's documentation.
Where it runs
The documented client runs the server on your machine.
INSTALLATION
Connect your client
You clone and build this one yourself.
Server setup
TERMINAL
git clone https://github.com/Nexus-Digital-Automations/Claude_Code_MCP_2
cd Claude_Code_MCP_2Install dependencies
TERMINAL
uv syncStart the MCP server
TERMINAL
python src/main.pyMCP Clients
Claude Desktop
CLAUDE DESKTOP · JSON
{
"mcpServers": {
"claude-code-mcp": {
"command": "uv",
"args": [
"--directory",
"<YOUR_PATH_TO_CLAUDE_CODE_MCP>",
"run",
"python",
"src/main.py"
],
"env": {
"PYTHONPATH": "<YOUR_PATH_TO_CLAUDE_CODE_MCP>/src"
}
}
}
}
