---
title: Getting started
description: Set up and deploy a working context layer your AI tools can read from and write to, in about fifteen minutes.
---

This guide helps you set up and deploy a working context layer your AI tools
can read from and write to. It takes about fifteen minutes. You need a browser
and the AI tool you already use, Claude or Codex.

Foundation comes in two ways. The hosted version runs on our infrastructure and is
the fastest way for you to experiment and play with it. If you would rather run it on your own infrastructure, everything
here applies once your deployment is up, and [Self-hosting](/self-hosting)
covers the setup.

## Create your workspace
Create a workspace from [midland.md](https://midland.md) and sign in with the
link sent to your email. There are no passwords. You are the workspace's first
admin: you decide what it holds, and you invite the team when you are ready.

## Connect it to your AI tool
Your workspace has one connector URL, shown when you sign in. It has the form
`https://midland.md/api/mcp`, and every host connects to it the same
way.

### Claude.ai
Install the connector once in Claude.ai and it is available in every Claude
surface on your account: the web, Claude Desktop, Cowork tasks and Claude
Code.

1. In Claude.ai, open **Settings**, then **Customize**.
2. Add a **Custom Connector**.
3. Name it `Midland` and paste your connector URL.
4. Leave **Authentication** and **OAuth client** as Claude detected them. Add
   no headers.
5. Follow the sign-in flow and approve access.
6. Restart Claude Desktop if it is open.

### Codex
Codex reads its connectors from one configuration shared by the desktop app,
the CLI and the IDE extension, so adding the workspace once makes it
available in all three.

1. Open **Settings** in the Codex desktop app, then **MCP servers**, then
   **Add server**.
2. Name it `midland`. The name becomes a key in Codex's configuration file,
   so keep it lowercase with no spaces.
3. Set the transport to **Streamable HTTP** and paste your connector URL.
4. Save, then choose **Restart** when Codex offers it.
5. Codex marks the server as needing sign-in. Choose **Authenticate** and
   approve access in the browser.

Any other MCP client connects to the same URL; see [Connector](/concepts/connector).

## Build the context layer
A workspace is only useful once it holds what your team knows, and the
quickest way to get there is to let your agent do the typing. Open a chat
with the connector attached and paste one of the prompts from
[Building context](/building-context). Each one starts from something you
already have (a website, a document or your AI tool's memory) and turns it
into shared context.

Nothing is saved until you agree to it, and you can see the result straight
away.

You do not need a team to start. Foundation is built for trying things on your
own first: fill a workspace with what you know, ask your agent questions
against it, and see whether the answers change. Invite people once they do.

## Next
- Invite a colleague. Their agent reads what yours wrote.
- Keep a number or a list current without a person: [Routines](/concepts/routines).
- See what teams build on it: [Examples](/examples).