Publish your tools as an MCP server

The tools you already built in NappAI, with their credentials, can be offered to other AI assistants as an MCP server. That way, from Claude, Cursor or Visual Studio Code you can look things up in your CRM, your warehouse or your databases without setting anything up again.

1. Build the flow

1. Add the MCPServer component.
2. Connect to its Tools input the tools you want to publish. It can be any component in tool mode, a Python Code or a whole agent connected through its Tool output.
3. Check each tool’s name and description: that is what the external assistant reads to decide when to use it.

Only the tools connected to MCPServer are published, and a flow has a single server.

2. Deploy a version

The MCP server always uses the deployed version of the flow. Deploy a version from the right bar of the editor. If you change the flow later, deploy another one so assistants see the changes. See Versions and publishing.

3. Connect it to your assistant

1. In the editor, press API Access and open the MCP Server tab.
2. Choose an API Key. See API keys.
3. Use the matching tab:

Tab	What it gives you
General	The MCP Server URL, with the key included, for any compatible assistant.
VS Code	An Install button that adds the server to Visual Studio Code, or the configuration to do it by hand.
Claude	The configuration for Claude Desktop and the command for Claude Code.
Cursor	An Install in Cursor button, or the manual configuration.

Set your assistant to the Streamable HTTP connection type, as the window says.

From then on, the assistant sees your tools, with their name and description, and uses them when it needs them. Each use is a run of the deployed flow, which you can review in Executions.

The key gives access to your tools

Anyone with the address and your API key can use the published tools, with your credentials. Share it only with whoever needs it, and use a separate key for each use, so you can withdraw it without affecting the rest.

Example

Central Hardware publishes its warehouse stock lookup. It builds a flow with MCPServer and a Python Code in tool mode that looks up the inventory, deploys a version and adds the server to Claude. From then on, anyone on the team asks Claude how many drills are left, and Claude gets the answer from the warehouse through NappAI.