Introducing Atlas App Connections: Secure, One-Click Access for AI Tools

For builders using AI coding tools, speed is everything. Integrations that require generating a credential, filing an access request, or waiting on a custom build can stop momentum (and productivity) in its tracks.

Until now, connecting an external application like an AI coding assistant to a MongoDB Atlas account introduced some of that friction. The only way to do so was by sharing long-lived service account credentials, a workaround that slowed builders down and left administrators managing a security risk they were reluctant to accept. Faced with that tradeoff, some organizations avoided third-party integrations, or defaulted to whatever tool was already the default in their AI stack, whether or not it was the best fit.

MongoDB Atlas App Connections closes that gap. It enables builders to securely connect AI coding tools to their MongoDB Atlas account with a single click. Built on OAuth 2.1, an industry-standard authorization framework, App Connections allows AI clients to act on the user's behalf via the MongoDB Atlas MCP Server, without long-lived credentials to generate, share, or manage.

Available now in Atlas App Connections

With Atlas App Connections, builders can connect their AI coding tools to MongoDB Atlas through the Atlas MCP Server, including:

• ChatGPT

• Claude

• Claude Code

• Codex CLI

• Cursor

• Devin

• Devin Desktop

• Grok

• Grok CLI

Once connected, these tools can provision, manage, and query MongoDB Atlas resources directly on the builder's behalf, whether that means spinning up a new cluster, checking on an existing one, or working with the data inside it, based on the access the connecting user already has. Administrators control how far that access extends, choosing between read-only and full access before any AI client can act.

Using App Connections in practice

Here's what that looks like for a builder, using Claude as an example. Assuming Claude is already running, connecting a MongoDB Atlas account takes three steps:

1. Install the MongoDB plugin by running /plugin install mongodb-atlas@claude-plugins-official.

2. Grant Claude access by running /mcp, selecting plugin:mongodb-atlas:mongodb-atlas, and choosing Authenticate.

3. Approve the one-click authorization.

Installing the plugin also adds a set of skills that teach Claude how to work with MongoDB Atlas, and installs MongoDB's remote MCP server behind the scenes. From there, the developer can simply ask Claude to provision a new database cluster, and Claude does it: translating that request into secure Atlas API calls through the Atlas MCP Server. There's no credential to generate or share to make that happen. The administrator overseeing that Atlas organization still has full visibility into the connection and can revoke it at any time.

What you get if you’re building

For builders, speed and security have always felt like a tradeoff. Bottlenecked by integration setup friction, many resorted to workarounds such as sharing high-risk, long-lived credentials just to keep building.

Atlas App Connections removes that tradeoff. A single click securely authorizes AI coding tools to connect to Atlas. There are no credentials to share and no manual configuration to manage. Builders stay in their flow, building at the speed of AI, without ever compromising account security to get there.

What you get if you’re governing an organization

For administrators, the calculus has been just as fraught. Engineering teams need modern AI coding tools to stay competitive, but granting those tools access via opaque, shared service accounts can mean losing track of who or what is actually acting on an organization's data.

Atlas App Connections eliminates that challenge. Access to connect AI clients to Atlas is disabled by default in existing organizations, and administrators must opt in. New organizations created during sign-up start with that access enabled, so first-time users get the best possible AI experience from the outset. Once approved, every action is attributed to both the application and the person who authorized it, visible in Atlas activity feeds. Administrators decide exactly how long an AI client's access can last before it requires re-approval, and can turn off AI client access for the entire organization at any time. Individual users can revoke their own connections whenever they choose. Atlas App Connections helps eliminate shadow AI without standing in the way of engineering teams doing their jobs.

How Atlas App Connections works

Connecting an AI coding tool to Atlas using App Connections takes five steps:

1. The user installs MongoDB's plugin or connector for their AI client.

2. The user requests to connect their MongoDB Atlas account from within that client.

3. The client redirects the user to Atlas.

4. The user logs in and sees a consent screen naming the application and exactly what it's asking to access.

5. All the user has to do is click “Approve” and the application is enabled to act on their behalf through the Atlas MCP Server.

Behind the scenes, that approval gives the AI coding tool permission to obtain a short-lived access token and a refresh token on the user's behalf, so it doesn't need to ask the user to log in again for every action. The tool uses that access to provision, manage, and query Atlas resources through the Atlas MCP Server, scoped to whatever access the connecting user already has. All of this happens automatically, invisible to both the user and the application.

A few properties of that access are worth knowing:

• User-scoped: The AI client can only take actions for which the user has permissions. If a user's Atlas permissions change, the AI client's access automatically changes with them.

• Revocable: Users can revoke an AI client's access at any time, and it takes effect within 15 minutes.

• Auditable: Every delegated action is attributed to both the AI client and the user it acted on behalf of.

• Governed by the organization: Delegated access to AI clients is disabled by default. Atlas organization owners decide whether to turn it on, and for AI clients specifically, whether to allow read-only or full access. If an owner turns off delegation for the entire organization, that change takes effect immediately.