MCP (Model Context Protocol)

MCP allows you to extend your agent’s capabilities by connecting it to external tools and services. An MCP server provides your agent with access to databases, APIs, file systems, or any custom functionality you want to expose.

Who this is for: Advanced users who want agents to call external tools, services, or internal systems.

What is MCP?

Model Context Protocol is an open standard for connecting AI agents to external tools. When you configure an MCP server, your agent can:

Call functions provided by the server (e.g., query a database, search the web, read a document).
Access data sources that aren’t part of the default agent capabilities.
Interact with your own internal tools and services.

Configuring MCP Servers (Desktop)

MCP configuration is currently available in the desktop application and is aimed at advanced users.

For most users, MCP is managed from:

Settings > Extensions > MCP Servers for user-level MCP servers.
The agent editor’s Advanced tab for agent-specific visibility and disablement.
Workspace configuration for workspace-level visibility.

Some builds may also expose a dedicated MCP configuration screen for editing raw JSON configuration directly.

Editing Raw Configuration

When the raw MCP configuration editor is available, it lets you manage MCP config as JSON. A typical config looks like this:

{
  "mcpServers": {
    "my-server": {
      "command": "npx",
      "args": ["-y", "my-mcp-server"],
      "env": {
        "API_KEY": "your-api-key"
      }
    }
  }
}

Each entry in mcpServers defines a server by its ID. Common fields include:

Field	Description
`command`	The command to run the server (for stdio transport).
`args`	Command-line arguments.
`env`	Environment variables to pass to the server.
`url`	URL for HTTP/SSE transport servers.
`transport`	Transport type (`stdio`, `sse`, `http`). Default: `stdio`.
`headers`	HTTP headers for URL-based transports.

Edit the JSON in the textarea and click Save.

Probing a Server

You can test an MCP server configuration before using it:

Enter the server ID (the key name in mcpServers) in the probe field.
Click Probe.
The result will show the server’s metadata, including available tools and their descriptions.

This is useful for verifying that a server is correctly configured and reachable.

Effective Runtime Configuration

The Effective Runtime MCP section shows what MCP servers are actually available in a given session after personal, workspace, and plugin configuration are merged together.

Enter a thread or session ID if needed.
Click Load Effective.
Review the merged result that will be available to the agent.

MCP in Agent Configuration

You can also manage MCP availability from other parts of the desktop app:

Per-agent: In the agent editor’s Advanced tab, disable specific MCP servers for that agent.
Per-workspace: In workspace configuration, review which MCP servers are active for that workspace.

Common MCP Servers

Here are some examples of MCP servers you might configure:

File system — Give your agent access to specific directories.
Database — Connect to PostgreSQL, SQLite, or other databases.
Web search — Enable web search capabilities.
Documentation — Provide access to your internal documentation.
Custom APIs — Wrap your internal services as MCP tools.

Security Considerations

MCP server configurations (including commands and environment variables) are stored locally on your desktop device.
When you connect to external MCP servers, data is sent to those servers and is governed by their privacy practices.
Be cautious when adding MCP servers from untrusted sources — they execute code on your machine.
Review the tools a server provides (via the Probe feature) before enabling it for an agent.

Next Step

Go to Extensions if you also need skills, hooks, plugins, or commands.
Go to Agents to restrict which MCP servers a specific agent can use.