Claude Desktop + Ollama: How to Run Open Models Locally and in the Cloud

Claude Desktop has traditionally meant one thing: use Anthropic’s models through Anthropic’s infrastructure.

That changed on August 25, 2026.

Ollama has introduced native support for connecting Claude Desktop with Cowork to Ollama as a third-party gateway provider, allowing developers to use models available through Ollama including local open models and Ollama cloud models without abandoning the Claude Desktop experience.

And the interesting part?

You no longer need to build an MCP bridge, manually edit configuration files, or wire together a complicated proxy.

Ollama can configure the connection for you.

Let’s set it up from scratch.

Think about the architecture like this:

Claude Desktop remains the interface. Ollama becomes the model gateway.

The model itself can be:

Ollama explicitly supports using any model available through Ollama, including both local and cloud models.

That creates an interesting developer workflow:

There are several practical reasons.

You can use an open model directly on your own machine instead of relying exclusively on a hosted proprietary model.

That can be useful for experimentation, private development workflows, offline work, or simply learning how different models behave.

Local hardware has limits.

A laptop might handle one model comfortably but struggle with a much larger reasoning or coding model.

Ollama lets Claude Desktop switch to models available through its cloud infrastructure as well.

You don’t need to learn another chat client just because you want to test open models.

Claude Desktop remains the UI.

Ollama handles the model connection underneath.

The new setup is surprisingly short.

Download and install Ollama for your operating system. Ollama supports macOS, Windows, and Linux.

After installation, verify that Ollama is available:

You should get the installed Ollama version.

Download and install the latest Claude Desktop application.

Claude Desktop is available for macOS and Windows.

Launch Claude Desktop once installation finishes.

Start the Ollama application.

The important part of the new integration is that you don’t have to manually configure Claude Desktop first.

According to Ollama’s current documentation:

That’s it.

This is considerably simpler than the older approach where developers commonly had to configure local integrations manually.

Inside Ollama, open the Claude integration and enable it.

When Claude is turned on, Ollama configures Claude Desktop to use Ollama as the gateway provider.

Conceptually:

Once enabled, Claude Desktop can use models configured through Ollama.

This is where the setup becomes interesting. Ollama allows you to configure the model Claude uses for each model option.

You can select a local model or an Ollama cloud model from the Ollama settings.

For example:

Your choice depends on what you’re trying to do. 

For quick experimentation, a local model can be enough.

For heavier coding, reasoning, or research workloads, a cloud model may make more sense.

Ollama has also continued expanding its cloud and coding-model support, including models designed for agentic development workflows.

Now open Claude Desktop and start a conversation.

The important thing to remember is that Claude Desktop is now acting as the client interface while Ollama provides the model gateway.

You don’t need to switch applications every time you want to test another model. That means your workflow can look like:

For developers, this creates a convenient environment for comparing model behavior while keeping the interaction layer familiar.

This is an important distinction.

Anthropic’s Model Context Protocol (MCP) is designed to connect AI applications with external tools, data sources, and services.

Claude Desktop supports local MCP servers, and modern Claude Desktop versions also provide desktop extensions for easier MCP installation.

But you don’t need MCP just to connect Claude Desktop to Ollama anymore. Ollama’s new integration works as a third-party gateway.

So don’t confuse these two architectures:

versus:

They solve different problems and you can potentially use both together.

Yes. This is one of the nicest parts of the integration.

Ollama says that you can turn the Claude integration off inside Ollama to restore your previous Claude setup.

So your workflow can become:

You’re not permanently replacing Claude. You’re adding another model path.

This is one of the most important questions whenever local AI enters the conversation.

Ollama states that telemetry is disabled by default and says it maintains a Zero Data Retention policy across its models and services, including local and cloud access.

However, developers should still understand the difference between local inference and cloud inference.

With a local model:

With a cloud model:

So don’t treat “Ollama” as automatically meaning every workload stays on your machine.

Know where the selected model runs before sending sensitive data.

This integration makes an interesting model-switching workflow possible.

Use a smaller local model for:

Use a larger cloud model for:

And use Anthropic’s models when you specifically want your normal Claude experience.

That gives you something many AI developers want: One interface, multiple model backends.

There’s another interesting part of the current Claude Desktop experience: Claude Cowork.

Cowork is designed for more hands-on computer work letting Claude work with files, organize information, and execute multi-step tasks on your desktop rather than acting only as a conversational assistant.

That makes the Ollama integration even more interesting for developers. Think about the stack this way:

Claude Desktop becomes the workspace, while Ollama provides another model layer underneath it.

That opens up a broader developer workflow.

Instead of using Cowork only with Anthropic’s models, developers can experiment with Ollama-supported models while keeping the same desktop environment for working with files and project-related tasks.

For example, you could imagine a workflow like:

This separation is important.

Cowork is the work environment. Ollama is the model gateway.

They are solving different problems, and that makes the combination potentially more flexible than thinking of Ollama simply as “another chatbot.”

A developer’s AI workflow rarely consists of just asking questions.

You might need to:

That’s where Cowork changes the equation.

You get a more task-oriented desktop experience, while Ollama gives you another way to choose the model powering the workflow.

The result is closer to:

And as the model ecosystem continues to expand, that separation between workspace, tools, and models becomes increasingly important.

What’s happening here is bigger than simply “Claude now works with Ollama.”

We’re moving toward a world where the AI application and the AI model are becoming separate layers.

Claude Desktop can be the application layer. Ollama can become the model gateway. MCP can provide the tool and data layer and the model can change underneath all of it.

That’s a powerful architecture for developers experimenting with the rapidly changing AI model ecosystem.

You don’t necessarily need to replace your favorite AI application every time a new model appears.

Change the model layer instead.

The old way of experimenting with different AI models often meant switching tools, clients, APIs, and configuration files.

Claude Desktop + Ollama reduces much of that friction.

When you want your normal Anthropic setup again, switch the Ollama integration off.

For developers, that’s the real value here: model flexibility without sacrificing the workflow you’re already comfortable with.

I hope you found it helpful and informative. If you have any questions or feedback, feel free to leave a comment below. Your support and engagement mean a lot to me.