Agents provision worktrees, not just query them
The MCP server exposes the full GTL toolset: create worktrees, allocate resources, wire up cross-repo dependencies, and control the dev server. AI agents work with real allocations instead of guessing ports and parsing stdout.
MCP: first-class automation interface
gtl mcp starts the Model Context Protocol server. Agents in Cursor, Claude Code, Windsurf, and other MCP-compatible editors call tools directly instead of parsing shell output.
- Provisioning —
new,setup,link,unlink,env_sync,config_set - Query —
status,list,port,routes,resolve,env,where,doctor,db_name,config_get - Supervisor —
start,stop,restart - Resources —
gtl://allocations,gtl://config/user
JSON flags for shell-only agents
gtl status --json— full allocation detailsgtl routes --json— routing URLs per portgtl doctor --json— health diagnosticsgtl env --json— env file contentsgtl resolve <project> --json— another project's URLgtl start --await— blocks until TCP accepts (with--await-timeout)
Agent context file
gtl install writes an AGENTS.md file with instructions for setup, teardown, and server management. Any MCP-compatible editor reads it as context. Cross-repo service URLs follow the same gtl resolve contract as humans.