You’re running agents in the GitHub Copilot app – maybe three at once, each on its own branch, each working a task while you steer. It’s a good way to work, right up until an agent needs to reach outside your repo.
It needs to read a SharePoint library, file a Salesforce record, or check an Outlook calendar, and suddenly it’s blind. An agent is only as capable as the tools you hand it.
The usual way to hand it one of those tools is to wire up a Model Context Protocol (MCP) server by hand: find the endpoint, paste it into config, add an auth header, keep the token alive, and hope you got the casing right. Then your teammate does the same thing on their machine, and the next person after that.
The MCP Connectors canvas removes the hand-wiring. It’s a plugin for the GitHub Copilot app: it lists the hosted MCP servers already published in your Azure Connector Namespace, and you connect one to Copilot by selecting it – no URL, no header, no local proxy.
Where the servers come from
Connector Namespace is the managed Azure service on the other end. It provides a list of curated MCP Servers that you can use to quickly create and manage MCPs connections. It also does the work you’d rather not: it stores and rotates the credentials, applies retry and throttling policies, and scales the server. Your machine never holds a raw secret related to the end systems that MCP touches, and access is governed by access policies, allowing fine grained control on who can use the MCP servers.
So, when you connect a server from the canvas, you’re not standing up infrastructure. The server already exists, already authenticated, already managed. You’re pointing Copilot at it.
Prerequisites
- GitHub Copilot app with canvas extension support.
- An Azure subscription with permission to cerate and view the namespace and to create connections and hosted managed MCP server configurations.
- Connector Namespace is in preview, and availability varies by region.
Install
From the GitHub Copilot app, add the following prompt.
Install connector-namespaces v1.2.0 for my user account, reload it, verify it is running, then open the MCP Connectors canvas.
Open the Connector Namespace canvas in a new session
Once you install if you need to open the Connector Namespace in a new session, just request it in a prompt:
Open the MCP Connectors canvas.
Connect an MCP server
- Open the MCP Connectors canvas.
- Select Sign in to Azure, then choose a subscription and Connector Namespace.
- Browse or search the MCP servers grouped under Microsoft and Partners.
- Select Connect and complete the connector’s separate authentication or consent flow.
- Confirm the server appears under My MCPs.
- Restart GitHub Copilot so a new session loads the added tools.
ℹ️Note
That last step matters more than it look. Copilot loads an agent’s tools when a session starts, so a server you add mid-session shows up in the next one, not the one you’re in.
Manage connected servers
My MCPs shows the servers connected to Copilot.
- Sandbox opens a connected server in the Connector Namespace playground, so you can try a tool call yourself before you let an agent lean on it.
- Disconnect removes the MCP registration from your MCP configuration. The MCP is still available in Connector namespace to be added again later. Also if you are sharing the connector namespace with other developers, this only remove the your local configuration.
- Connect adds configured MCP connectors that are not registered locally yet to your local MCP configuration.
- Delete from namespace… removes the MCP registration from your MCP configuration and deletes the MCP connection from your connector namespace. If you are sharing the connector namespace with other developers this will affect any developer using this MCP connector – be careful when using this.
- Switch namespace switches the active subscription or namespace – useful when you keep dev and production connections apart.
How it works and security
The extension registers each server directly in Copilot’s user-scoped MCP configuration. There’s no local MCP proxy sitting in the tool path between the agent and the server.
The canvas itself runs on loopback. Your Azure tokens stay in process memory, the selected namespace coordinates may be cached locally, and gateway credentials are sent only to the server’s configured HTTPS endpoints. The connector’s own secrets never reach your machine. That’s the namespace’s job, which is also why an admin can rotate or revoke a connection in one place instead of chasing down every developer’s config.
Where can I use my connected MCP servers
You manage the servers from the canvas, but they don’t only work there. The canvas writes to Copilot’s user-scoped MCP configuration at ~/.copilot/mcp-config.json, and the GitHub Copilot app is built on GitHub Copilot CLI. Connect a server once and it’s available to both.
- Agent sessions in the app. Any session on your machine picks the server up, whatever repository or branch it’s working on, because the config is user-scoped rather than per-project.
- Copilot CLI. The same server shows up in the terminal. Run copilot mcp list to confirm it, or /mcp show SERVER-NAME in an interactive session to see the tools it exposes.
Two things to know:
- Tools load when a session starts, so connect first and start the session second.
- A project-level .mcp.json or .github/mcp.json takes precedence over your user config when the names collide, which is what you want if a repository pins its own version of a server.
VS Code is the exception. Copilot Chat there reads .vscode/mcp.json, a separate file the CLI doesn’t use, so a server you connect in the canvas won’t appear in VS Code until you configure it there too.
Where this all sits today
The GitHub Copilot app went generally available in June 2026, and canvases shipped with it, so neither is a preview feature. The MCP Connectors canvas is a published extension you install at personal scope. You get the version we ship and tested, not one an agent generates on the fly from a prompt.
Two caveats. GitHub doesn’t publish a compatibility guarantee for the canvas extension format, so an app update may need a matching update from us. Re-run the install commands above to pick it up if you see any issues. On the Azure side, Connector Namespace is still in preview. Check the documentation for more details on this service outside the GitHub Copilot App support.
If you’re on Copilot Business or Enterprise, check with your admin that the GitHub Copilot app policy is enabled. It’s on by default, and it’s separate from the Copilot CLI policy.
Try it
Install the plugin, point it at a namespace, and connect one server your agents keep reaching for: the SharePoint library, the CRM, whatever it is. Then watch an agent actually use it in the next session. Tell me where the canvas saved you a config headache and where it got in your way. That’s the feedback we act on before general availability.