Agent Control via MCPCopy Anchor Link

devsy mcp serve runs a Model Context Protocol server over stdio. MCP-compatible AI clients can connect to it and drive Devsy - listing, creating, and executing inside workspaces - without shelling out to the CLI themselves.

devsy mcp serve

The server reads JSON-RPC frames from stdin and writes them to stdout. Configure your MCP client to launch the command; it will discover the registered tools at startup.

Tools exposedCopy Anchor Link

The server registers 11 tools across workspace and provider domains:

workspace_exec is bounded by three server flags:

• --exec-timeout-default (default 5m) - default per-call timeout.
• --exec-timeout-max (default 30m) - caller-supplied timeouts are clamped to this.
• --exec-output-cap (default 100 KiB) - per-stream byte cap; excess is replaced with a truncation marker.

Example client configCopy Anchor Link

The exact format depends on your MCP client, but every client needs the command to launch and any arguments. A generic JSON shape looks like:

{
  "mcpServers": {
    "devsy": {
      "command": "devsy",
      "args": ["mcp", "serve"]
    }
  }
}

Check your client's documentation for where this config lives (e.g. claude_desktop_config.json for Claude Desktop). Refer to the MCP specification for transport and tool semantics.

Install the Devsy Agent SkillCopy Anchor Link

MCP provides the tools to control Devsy. The first-party Agent Skill teaches an agent how to select workspaces and providers, execute commands, recover from timeouts, and use CLI diagnostics when needed. It is for using Devsy, rather than contributing to the Devsy source repository.

After installing Devsy, configure your agent's MCP client to launch devsy mcp serve, then install the skill:

npx skills add devsy-org/devsy --skill devsy

For use across projects, add -g. To target a particular agent, use --agent:

npx skills add devsy-org/devsy --skill devsy -g
npx skills add devsy-org/devsy --skill devsy --agent codex
npx skills add devsy-org/devsy --skill devsy --agent claude-code

The Agent Skill and MCP configuration are separate. Installing the skill does not install Devsy, configure MCP, or set up a provider. The installer requires Node.js/npx; Devsy itself does not depend on it. See the skills installer for supported clients and updates. Follow your agent's instructions for reloading newly installed skills or MCP configuration.

Hand this to your agentCopy Anchor Link

Copy this prompt into the agent you want to use with Devsy:

Set up the published first-party Devsy skill for this agent so you can operate
Devsy workspaces for me. Check whether Devsy is installed and supports
`devsy mcp serve`. Install the devsy skill from devsy-org/devsy for this agent;
use a project installation unless I ask for global installation. Follow this
agent's current MCP setup instructions to configure Devsy, preserving existing
MCP entries. If installation or configuration cannot be completed, tell me
which prerequisite is missing and the exact next step. Verify the integration
by listing my Devsy workspaces without creating or changing any. Report what
was installed/configured and whether the connection worked.

After setup, try: "Create a Devsy workspace for this repository and run its tests," or "Stop my example workspace." The skill prefers MCP tools and uses the CLI for operations the connected server does not expose. For live running state, the current MCP workspace_status result is only saved configuration; devsy workspace status <name> --result-format json probes live state.

SecurityCopy Anchor Link

The MCP server has the same authority as your local Devsy CLI. A connected client can create, modify, and delete workspaces, run arbitrary commands inside them, and add or remove providers. Only enable devsy mcp serve for clients you trust.