This project exposes a local development workspace over MCP so ChatGPT, Claude, or another MCP-capable host can operate directly on this machine's approved development directories.
The goal is not to delegate work to a separate local coding agent. The MCP host should call tools that read files, edit files, search code, and run shell commands directly against approved local project roots.
Pi's SDK is currently used as the backend adapter for mature local coding primitives such as read, edit, write, grep, find, ls, and bash. DevSpace wraps those primitives behind a remote Streamable HTTP MCP interface, suitable for use through a Cloudflare Tunnel.
The model-facing workflow is workspace based. MCP clients should call
open_workspace once per local project directory or worktree, then reuse the
returned workspaceId for subsequent tool calls in that same folder. Do not
call open_workspace again for the same folder unless the workspaceId is
rejected as unknown, the client switches folders/worktrees or checkout/worktree
mode, or the user explicitly asks to reopen. AGENTS.md files are returned
automatically by open_workspace and by later tool calls when the requested path
enters a directory with instructions that have not been loaded for that
workspace.
Core constraints:
- Treat this as remote access to the local machine; security is part of the core design, not a later add-on.
- Start with a narrow filesystem allowlist.
- Prefer explicit, inspectable tool calls over autonomous local agent loops.
- Keep the first version small enough to validate with real ChatGPT/Claude MCP clients before adding UI or workflow features.