The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the Cloudgate Workflow listing page.
Build and manage Cloudgate workflow-APIs — controllers, actions, workflow graphs, and databases — from any MCP-compatible AI client, including Claude, OpenAI Codex, and others.
Cloudgate runs a hosted, remote MCP server at api.cloudgate.dev. It speaks the standard
Model Context Protocol over streamable HTTP and authenticates each user with OAuth, so any
client that supports remote MCP servers can connect. This repo also packages the server as a
ready-to-install Claude plugin for convenience.
https://api.cloudgate.dev/mcp/workflowdev.cloudgate/workflowhttps://github.com/cloudgatedevs/mcphub.cloudgate.dev to sign in, then the tools become available.Add the remote server to ~/.codex/config.toml:
Then authenticate:
Any client that supports remote (streamable HTTP) MCP servers can connect directly to
https://api.cloudgate.dev/mcp/workflow and complete the OAuth sign-in.
The packaged plugin uses native HTTP and requires a client with remote MCP OAuth support.
It does not launch Node.js or mcp-remote. The previously unpinned bridge could update
independently of the plugin; version 0.10.0 drops the authorization callback issuer when
calling its SDK. Do not replace this configuration with an unpinned bridge command.
The Cloudgate MCP server uses OAuth 2.1 (authorization code + PKCE). Each user signs in as themselves; every MCP call is scoped to that user's tenant and account. No tokens or secrets are stored in this repository.
The hosted server works on its own with any MCP client. This repo additionally ships the
cloudgate-builder Claude plugin, which bundles the connection config plus a
cloudgate-build skill — a playbook that guides the model through creating controllers,
actions, workflow graphs, and databases correctly.
© Cloudgate Dev LLC.