Set up a coding agent
Set this up when a coding agent works in a repository that runs on devlanes. Your agent can do it for you:
Install the devlanes skill:npx skills add devlanes-run/devlanes --skill devlanes
Then check that your shell can run devlanes in this repository.If it cannot, register the devlanes MCP server with this agent.Tell me which steps I have to do myself, such as restarting you.Install the skill
Section titled “Install the skill”The skill teaches the agent to run dev servers and watchers through the daemon, so they outlive its shell commands, and to test and stop what it started.
$ npx skills add devlanes-run/devlanes --skill devlanesRegister the MCP server
Section titled “Register the MCP server”The MCP server offers devlanes commands to the agent as tools. Register it when the agent host has no shell, or when its sandbox keeps the shell away from the daemon.
Claude Code
$ claude mcp add --scope user devlanes -- devlanes mcpCodex
$ codex mcp add devlanes -- devlanes mcpOpenCode
$ opencode mcp add devlanes --global -- devlanes mcpCursor
{ "mcpServers": { "devlanes": { "type": "stdio", "command": "devlanes", "args": ["mcp"] } }}Each registers the server for every project. devlanes mcp --help prints the lines for other hosts.
If the agent should only read
Section titled “If the agent should only read”Add --read-only after mcp, and the agent gets the tools for status, logs and errors only:
$ claude mcp add --scope user devlanes -- devlanes mcp --read-onlyIf the agent’s shell is sandboxed
Section titled “If the agent’s shell is sandboxed”Use devlanes from an agent sandbox says which way in works.