Skip to main content
Model Context Protocol (MCP) is an open standard that lets AI assistants call external tools provided by third-party servers. By connecting an MCP server to Luminy, you instantly give the agent new capabilities — web search, GitHub access, database queries, cloud provider APIs, or any custom tool you build — without any changes to Luminy itself. Once a server is connected, its tools appear in the agent’s palette automatically and are used just like built-in tools.

​
Supported transports

Luminy supports two ways to connect an MCP server:

stdio

Luminy spawns a local process on your machine and communicates over stdin/stdout. Ideal for lightweight tools distributed as npm packages or executables.

HTTP / SSE

Luminy connects to a running server over HTTP using Server-Sent Events. Use this for hosted or remote MCP servers.

​
Adding an MCP server

1

Open Settings → MCP

Click the settings icon in the sidebar, then select the MCP tab.
2

Click Add Server

Press the + Add Server button to open the connection dialog.
3

Choose a transport

Select stdio for a local command or HTTP for a remote server URL.
4

Enter the connection details

For stdio: Enter the executable in the Command field and any arguments in the Args field (one per line). For example, to add the Brave Search server:
  • Command: npx
  • Args: -y then @modelcontextprotocol/server-brave-search (one per line)
You can also set Environment variables as KEY=VALUE pairs (one per line) if the server requires them.For HTTP: Enter the server URL in the URL field, for example:
https://mcp.example.com/mcp
When you type an HTTP URL, Luminy automatically probes the server and fills in the name if it exposes a discovery endpoint.
5

Name the server

Give the server a human-readable name. Luminy may pre-fill this from the server’s discovery metadata.
6

Save

Click Add. Luminy connects to the server immediately and lists the tools it exposes. If the connection succeeds, the server’s tools are available to the agent right away.

​
OAuth authentication

Some MCP servers — typically hosted services — require OAuth authorization before they will accept requests. Luminy handles the full OAuth 2.0 PKCE flow for you:
  1. Luminy detects the OAuth requirement automatically when you enter the server URL.
  2. A browser window opens for you to sign in and grant access.
  3. After you authorize, Luminy receives the tokens and stores them securely. The server is added and connected.
If a server requires static client registration (pre-issued client credentials not obtained via dynamic registration), it is not currently supported. Luminy only supports servers that implement dynamic client registration.

​
Example servers to try

These are all stdio servers. For each one, set Command to npx and enter the remaining tokens as separate lines in Args.
ServerArgs (one per line)What it adds
Brave Search-y / @modelcontextprotocol/server-brave-searchWeb search results
Filesystem-y / @modelcontextprotocol/server-filesystem / /path/to/dirScoped read/write access to a directory
GitHub-y / @modelcontextprotocol/server-githubRepository access, issues, pull requests
Set any required API keys (such as BRAVE_API_KEY=your_key) in the Environment field.
You can also import MCP server configs in bulk by pasting a JSON object in the standard mcpServers format. Luminy parses the object, adds each entry, and attempts to connect them all at once.

​
Managing servers

All connected servers are listed on the Settings → MCP page. From there you can:
  • Enable / disable a server with the toggle. Disabling a server disconnects it without deleting its configuration.
  • Test the connection to refresh the tool list and verify the server is reachable.
  • Remove a server to disconnect it and delete its configuration permanently.

​
Important notes

Only connect MCP servers you trust. An MCP server’s tools run with the same permissions as any other agent tool. A malicious server could instruct the agent to read files, run commands, or make network requests on your behalf. Treat MCP servers the same way you would treat any third-party CLI tool you install.
stdio servers run as child processes on your machine. They inherit your user environment and have access to your filesystem paths. HTTP servers communicate over the network; be mindful of what data you send to remote endpoints.