MCP in NappAI
MCP (Model Context Protocol) is an open standard that many apps use to offer ready-made tools for an AI agent: look something up in a CRM, create a task, search documentation, check a domain… Each app publishes its MCP server and any compatible assistant can connect to it.
In NappAI, MCP works both ways:
| Direction | What you get | Page |
|---|---|---|
| Use MCP servers | Your agents and flows use the tools another app publishes, without a dedicated component for it. | Use MCP servers |
| Publish your MCP server | Assistants such as Claude, Cursor or Visual Studio Code use the tools you built in NappAI, with your credentials. | Publish your tools |
This is how both are done, end to end:
What you see in the video
- An agent that suggests brand names gets, from the MCPs category of the palette, the Find-A-Domain server, which checks whether a domain is free.
- Its credential is created, its Tool output is connected to the agent’s tools and the
check_domaintool is chosen, with its data marked for the AI to fill in. - In the playground, the agent suggests five names and checks each domain with the MCP server.
- In another flow, a warehouse stock lookup is published with the MCP Server component. The API Access window gives the server address and the steps to connect it.
- An external assistant sees the published tool, uses it, and gets the answer from the deployed flow.
Related components
- MCP Connection: the component that connects to an MCP server.
- MCPServer: the component that publishes your tools.
- If you want your agents to work with agents on other platforms, see A2A.