Skip to main content
The MCP Server node can host its own MCP server. Wire integration nodes into it, pick the operations to expose, and NoClick mints a public URL that any MCP client can connect to. Calls run with the credentials connected in your workflow, so your assistant gets working Slack, Linear, or GitHub tools with zero setup on the client side.

​
Build a hosted server

1

Add an MCP Server node

Add an MCP Server node to the canvas and leave its Server URL field empty. An empty URL puts the node in hosting mode.
2

Wire in integration nodes

Drag from an integration node’s top handle into the MCP Server node’s bottom handle. You can wire in multiple nodes, each with its own credential.
3

Allowlist operations

Click a wired node. Its config panel switches to an operation allowlist; pick the actions the server should expose. Each becomes a tool named {provider}__{operation} (for example linear__create_issue), plus an automatic {provider}__lookup_options tool for resolving IDs.
Operation allowlist for a node wired into an MCP Server node
4

Copy the server URL

Select the MCP Server node and click Connect externally at the top of its config panel. The dialog mints your server URL and shows setup guides for each client:
https://mcp.noclick.app/s/{id}
Connect externally dialog showing the hosted MCP server URL
5

Connect from your client

Paste the URL into any MCP client that speaks Streamable HTTP. No authentication step is needed; the URL is the key.

​
Client setup

Claude (claude.ai and the desktop app): go to Settings > Connectors, choose Add custom connector, paste the server URL, and click Add. ChatGPT: turn on Developer mode under Settings > Apps & Connectors > Advanced settings, then click Create, paste the URL as the MCP Server URL, and set Authentication to No Authentication. Cursor: add the server to .cursor/mcp.json in your project (or ~/.cursor/mcp.json for all projects): 
{
  "mcpServers": {
    "noclick-my-server": {
      "url": "https://mcp.noclick.app/s/{id}"
    }
  }
}
The Connect externally dialog also has guides for Claude Code, Codex, VS Code, Windsurf, Zed, and a generic config for other clients.

​
Tools resolve live

The tool list is read from your workflow graph on every request. Adding a node, changing an allowlist, or swapping a credential applies on the client’s next tool listing, with no publish step. Draft workflows serve too.

​
Security model

The link is the capability. Anyone who has the URL can call the exposed tools as you, with the credentials connected in the workflow. Share it like a secret.
  • To revoke access, click Rotate next to the URL in the Connect externally dialog. The old link stops working immediately; update every client that used it.
  • Every call is audited, so you can review what was invoked and whether it succeeded.
  • Only the operations you allowlist are exposed. Keep the list minimal.

​
External mode

The same node has a second, mutually exclusive mode: set Server URL to an external MCP server endpoint and the node proxies that server’s tools to AI agents in your workflow instead. External mode supports Streamable HTTP and SSE transports, API key or OAuth authentication, a tool name prefix, and a tool filter. One node cannot do both. If it has a Server URL and wired tool nodes at the same time, it returns an error instead of picking a side. Use two MCP Server nodes if you need both.

​
Next steps

Agent tools

Wire the same integrations directly into an AI agent instead.

External MCP servers for agents

Give your agents tools from external MCP servers.