MCP Server

fluidgrids-mcp, from @fluidgrids/mcp-server, is a Model Context Protocol server that exposes FluidGrids to AI assistants such as Claude Desktop and Claude Code. Once connected, the assistant can list and create workflows, trigger and follow runs, read logs, browse nodes, and manage credentials, webhooks and schedules in your workspace, with your permissions.

Install

npm install -g @fluidgrids/mcp-server
# or run it without installing
npx -y @fluidgrids/mcp-server

It needs Node.js 24.11 or later (or Bun 1.3.14 or later). The server speaks stdio by default, for desktop clients, and HTTP with --http for a shared remote server:

fluidgrids-mcp                      # stdio
fluidgrids-mcp --http --port 4100   # HTTP, bound to 127.0.0.1

Configure a client

Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "fluidgrids": {
      "command": "npx",
      "args": ["-y", "@fluidgrids/mcp-server"]
    }
  }
}

Claude Code:

claude mcp add fluidgrids -- npx -y @fluidgrids/mcp-server

Sign in

Ask the assistant to sign you in. The fluidgrids_auth_login tool starts the device-code flow: it gives you a URL and a code to approve in your browser, and the token is then saved locally in ~/.config/fluidgrids-mcp/ and refreshed automatically. Then ask it to pick a workspace (fluidgrids_auth_select_workspace).

For an unattended server, pass tokens or an API key through the environment instead (see below). Create API keys under Developer Portal › API Keys in the app.

Environment variables

Variable	Purpose	Default
FLUIDGRIDS_WORKSPACE_ENDPOINT	Workspace gateway	https://graphqlworkspaces.burdenoff.com/workspaces/graphql
FLUIDGRIDS_GLOBAL_ENDPOINT	Global gateway	https://graphql.burdenoff.com/global/graphql
FLUIDGRIDS_OIDC_ISSUER	Device-code sign-in issuer	derived from the global endpoint
FLUIDGRIDS_ACCESS_TOKEN, FLUIDGRIDS_REFRESH_TOKEN, FLUIDGRIDS_WORKSPACE_TOKEN	Pre-issued tokens; skip the device-code flow
FLUIDGRIDS_API_KEY	Authenticate with an API key
FLUIDGRIDS_WORKSPACE_ID, FLUIDGRIDS_ORGANIZATION_ID	Preselect the workspace and organisation
FLUIDGRIDS_HTTP_PORT	Port for --http	3002

For alpha, set FLUIDGRIDS_WORKSPACE_ENDPOINT=https://alphagraphqlworkspaces.burdenoff.com/workspaces/graphql and FLUIDGRIDS_GLOBAL_ENDPOINT=https://alphagraphql.burdenoff.com/global/graphql.

What the assistant can do

Tool names start with fluidgrids_.

Area	Tools
Workflows	list, get, versions, create, update, delete, trigger, publish, activate, deactivate, duplicate, archive
Runs	list, get, logs, cancel, pause, resume, retry a node
Nodes	list, get
Credentials	list, get, list types, create, attach a secret, test, update, delete
Webhooks and schedules	list, get, create, update, delete, activate or pause
Dashboard and search	summary, usage, search
Platform	sign-in and account, workspaces, roles, tags, files, integrations, notifications, export

Two prompts ship with the server: fluidgrids-build-workflow (author a workflow step by step) and fluidgrids-debug-run (work through a failed run). Two resources give the assistant context: fluidgrids://schema and fluidgrids://nodes.

Safety

• Every tool call goes through the same API and permission checks as the app, so the assistant can only do what the signed-in user can do.
• Destructive tools (deleting or archiving a workflow, cancelling a run, deleting a credential or webhook, revoking keys or sessions) do nothing unless the call passes confirm: true. Without it they answer that the action was not executed, so the assistant has to ask you first.
• Tools carry MCP hints (readOnlyHint, destructiveHint), so clients that support them can ask before running a risky tool.

Next steps

• CLI Reference
• GraphQL API