Start free

Xano MCP Builder: build and host MCP servers, governed like everything you ship.

With the Xano MCP Builder, your coding agent writes MCP servers in TypeScript with the Xano SDK. Test them locally, then deploy them authenticated, permissioned, observable and versioned.

Start free Talk to sales Talk to an engineer

How AI code governance works

Write with AI · Run locally · Deploy to Xano

How to build an MCP server on Xano

Your coding agent writes it in TypeScript with the Xano SDK, you run it on the Xano Engine on your machine, then you deploy it to Xano, which hosts it.

  1. 01 · Write with AI

    Your coding agent writes the MCP server in TypeScript.

    Run npx @xano/sdk init my-app, then ask Claude Code, Codex or Cursor for the tools you want agents to have. The agent writes the MCP server, its tools, prompts and resources, and the tables and functions behind them as TypeScript objects with the Xano SDK, following the AGENTS.md that init wrote.

    $ npx @xano/sdk init my-app && cd my-app
  2. 02 · Run locally

    You run it on the Xano Engine, on your machine.

    npm run xano:deploy typechecks the code and runs the MCP server on the Xano Engine, a single binary with Postgres built in, with no account and no network. Test the tools against real data and run npx xanosdk test run-all before anything leaves your laptop.

    $ npm run xano:deploy
  3. 03 · Deploy to Xano

    You deploy it to Xano, where it is hosted for you.

    One command puts the same code in a disposable ephemeral on Xano’s cloud and runs its tests. Promote the release and the MCP server is live: authenticated, permissioned, observable and versioned, under your workspace rules.

    $ npx xanosdk deploy --ephemeral --test
  4. 04 · Connect

    Your agents connect and call the tools.

    Claude, Cursor, ChatGPT and anything else that speaks MCP connect to the hosted server. Every tool call lands in request history and audit logs, attributed to the agent that made it.

The Xano SDK, command by command

An MCP server is a release, not a side door

Build, govern, observe and connect MCP servers

Tools, prompts and resources your agents use ride the same track as every API, task and table: built, governed, observed, connected.

  1. 01 · Build

    Build tools agents can use

    Expose your data and logic as MCP tools, prompts and resources without standing up infrastructure. Your coding agent writes the MCP server as one more TypeScript object in your repo with the Xano SDK, and the Xano editor shows your team every tool, prompt and resource it exposes.

    MCP Builder docs Xano MCP overview
    table orders function refund MCP server hosted on Xano tool search_orders tool get_order tool issue_refund Claude Cursor ChatGPT your data and logic as tools · no infrastructure to stand up
  2. 02 · Govern

    Governed by default

    The same workspace rules apply: auth patterns, allowed dependencies, review gates. An MCP server is a release, not a side door.

    Security
    The release line an MCP server ships like everything else auth dependencies review side door closed
  3. 03 · Observe

    Observable

    Request history and audit logs on tool calls. When an agent acts, it's attributable and reviewable like any other change.

    Observability
    Tool calls Claude search_orders ✓ Cursor get_order ✓ ChatGPT issue_refund ✓ Claude get_order ✓ request history on every call Audit log ◆ which agent ◆ which tool ◆ what changed attributable reviewable like any other change
  4. 04 · Connect

    Connected

    Works with Claude, Cursor, ChatGPT, and anything that speaks MCP.

    Integrations
    Claude Cursor ChatGPT any MCP client MCP server hosted on Xano
Hosted on Xano

MCP server hosting: auth, permissions, logs and versions

Xano runs your MCP servers for you on the same Kubernetes infrastructure as your Postgres database, REST APIs, realtime channels, background tasks and AI agents. There is no separate infrastructure to stand up.

Authentication

MCP servers run behind the approved auth patterns your team sets once as workspace rules, the same way your APIs do.

Permissions

Allowed dependencies and review gates apply to MCP servers like any other release. Nothing reaches production without passing the gate.

Logs

Request history and audit logs on every tool call. When an agent acts, it is attributable and reviewable like any other change.

Versions

Every release is an immutable, version-controlled deploy object, in your cloud or ours, with instant rollback.

Questions

The Xano MCP Builder

What is an MCP server?

An MCP server gives AI agents three things through the Model Context Protocol (MCP), the shared language agents use to discover and use outside systems: tools they can call, prompts they can reuse and resources they can read as context. On Xano, the tools are reusable actions agents invoke within your backend, built on the same data and logic as your APIs.

How do I host an MCP server?

Deploy it to Xano and Xano hosts it for you. Your coding agent writes the MCP server in TypeScript with the Xano SDK, you test it on the Xano Engine locally, and one command deploys it to Xano’s cloud, authenticated, permissioned, observable and versioned, with no infrastructure to stand up.

Can Claude and Cursor call a Xano MCP server?

Yes. A hosted Xano MCP server works with Claude, Cursor, ChatGPT and any other client that speaks MCP. Each tool call is recorded in request history and audit logs and attributed to the agent that made it, so you can review what every agent did.

Is it free to try?

Yes. The Free plan includes the MCP and AI agent builder, server-side business logic and a PostgreSQL database with up to 100,000 records, at no cost. You can also build and test on the Xano Engine on your own machine with no account. Compare plans.

AI builds software. Xano makes it trustworthy.

One platform to build, run and govern AI-built software.

Start free Talk to sales Talk to an engineer