Updated October 2026
Use MCP servers in a desktop app with any model
The Model Context Protocol (MCP) is the open standard for connecting AI apps to tools and data. An MCP client is the app that connects to MCP servers and lets a model use them. CoWork OS is a free, open-source MCP client that works with whichever model you choose, including local models, and adds approvals so a tool call that changes something waits for you.
What CoWork OS supports
| Capability | Details |
|---|---|
| MCP client | Connect local (stdio) servers and remote servers over SSE, WebSocket or streamable HTTP, including OAuth sign-in |
| MCP registry | Browse and install servers from the official MCP registry inside the app |
| MCP host | Expose CoWork's own tools as an MCP server over stdio for other MCP clients |
| Secure MCP Tunnels | Reach a private MCP server through an outbound-only connection to a relay you operate (feature-flagged) |
| Any model | Cloud providers, gateways, Ollama or MLX |
Add a local MCP server
- Open Settings → MCP Servers and click + Add Server.
- Choose Local command.
- Fill in Server Name, Command (for example
npx -y @modelcontextprotocol/server-filesystem), Arguments and any Environment Variables (oneKEY=valueper line). - Save, then ask a task to use the server's tools.
Add a remote MCP server
Choose Remote (OAuth) after + Add Server, enter the server URL and sign in when prompted. Or use Browse Registry to pick a listed server.
From the command line
cowork mcp list
cowork mcp add
cowork mcp test
The CLI reference lists every option, and Secure MCP Tunnels covers private servers.
Why use an MCP client that isn't tied to one model
Many MCP clients are built around a single vendor's model. With CoWork OS the same MCP servers work whether a task runs on Claude, GPT, Gemini, an open model through a gateway, or a local model on your own machine, and each task's access profile decides which tool calls need your approval.
Frequently asked questions
What is an MCP client?
An app that connects to Model Context Protocol servers and lets an AI model call their tools. CoWork OS is an open-source MCP client for desktop that works with any supported model.
Can I use MCP servers with local models?
Yes. In CoWork OS, MCP servers work with local models through Ollama or MLX as well as cloud models, as long as the model supports tool calling well enough for the task.
Does CoWork OS support remote MCP servers?
Yes. It supports remote servers over SSE, WebSocket and streamable HTTP, including servers that use OAuth sign-in.
Can CoWork OS act as an MCP server?
Yes. Its MCP host mode exposes CoWork’s tools as an MCP server over stdio so other MCP clients can use them.
CoWork OS itself is free and open source. Model providers, gateways, connected services, and account plans may charge separately and apply their own eligibility and usage rules.
Last reviewed 2026-10-20 against official documentation. Features, plans and prices change; check the linked sources before deciding.