Skip to main content

MCP Server

fluidgrids-mcp 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 runs, read logs, browse nodes, and manage connections in your workspace — with your credentials and your permissions.

Install

The package is not published to npm yet; build it from source:

git clone https://github.com/algoshred/fluidgrids-mcp.git
cd fluidgrids-mcp
bun install
bun run build

The server supports three transports: stdio (default, for desktop clients), SSE, and HTTP (for a shared remote server).

node dist/index.js                          # stdio
node dist/index.js --transport sse --port 3232

Configure a client

Claude Desktop (~/.config/claude/claude_desktop_config.json on Linux, the equivalent path on macOS and Windows):

{
"mcpServers": {
"fluidgrids": {
"command": "node",
"args": ["/path/to/fluidgrids-mcp/dist/index.js"],
"env": {
"FLUIDGRIDS_GRAPHQL_URL": "https://graphqlworkspaces.burdenoff.com/workspaces/graphql",
"FLUIDGRIDS_API_URL": "https://graphqlworkspaces.burdenoff.com",
"FLUIDGRIDS_ACCESS_TOKEN": "your-token"
}
}
}
}

Claude Code: claude mcp add fluidgrids -- node /path/to/fluidgrids-mcp/dist/index.js with the same environment variables.

Get a token with fluidgrids auth login and fluidgrids auth status (see the CLI Reference), or create an API key under Developer Portal › API Keys in the app.

Environment variables

VariablePurpose
FLUIDGRIDS_GRAPHQL_URLWorkspace GraphQL gateway endpoint (all workflow, run, node, and credential tools).
FLUIDGRIDS_API_URLBase URL of the same gateway, used for the REST endpoints (webhooks, health).
FLUIDGRIDS_ACCESS_TOKENBearer token or API key. The workspace is the one the token was issued for.
FLUIDGRIDS_TIMEOUTRequest timeout in milliseconds (default 30000).
MCP_PORT, MCP_HOSTBind address for the SSE/HTTP transports.

What the assistant can do

AreaTools
Workflowslist, get, create, update, delete, trigger, publish
Runslist, get, get logs, cancel, retry
Nodeslist, get (with schema), list categories
Credentialslist, create, delete
AI assistancegenerate a workflow from a description, suggest nodes, diagnose an error
Otherhealth, sign in, dashboard stats, search, list and create webhooks

Three prompts ship with the server: workflow-builder (step-by-step workflow authoring), troubleshoot (work through a failed run), and optimize (suggestions for an existing workflow).

Everything the server does goes through the same GraphQL API and RBAC as the app, so the assistant can only do what the signed-in user can do.

Next steps