Skip to content

MCP Settings ​

Configure Model Context Protocol (MCP) servers to extend AI capabilities.

What Is MCP? ​

MCP (Model Context Protocol) is a standard for connecting AI models to external tools and data sources. MCP servers provide additional capabilities like file access, database queries, web browsing, and more.

Tabs ​

Marketplace ​

Browse and install MCP servers from the marketplace. The marketplace provides pre-configured servers that are ready to use.

Installed ​

View and manage your installed MCP servers. Configure settings, enable/disable servers, and remove unwanted servers.

Installing MCP Servers ​

From Marketplace ​

  1. Go to the Marketplace tab
  2. Browse available servers
  3. Click Install on the server you want
  4. The server will be downloaded and configured automatically

Manual Installation ​

For servers not in the marketplace:

  1. Go to the Installed tab
  2. Add server configuration manually
  3. Provide the command to start the server

Managing Servers ​

Enabling/Disabling ​

Toggle servers on or off. Disabled servers won't be started and their tools won't be available.

Configuration ​

Some servers may require configuration:

  • Environment variables
  • API keys
  • Connection settings

Removing Servers ​

  1. Click the remove button on a server
  2. Confirm the removal

Available Server Types ​

MCP servers can provide various capabilities:

  • File System: Read and write files
  • Database: Query databases
  • Web: Browse and fetch web content
  • API Integration: Connect to external services
  • Code Execution: Run code in sandboxes

Tool Output Safety ​

Alma can compact large MCP tool outputs to save tokens. MCP servers can mark tool results as exact or compact depending on how safely they can be truncated.

Supported modes:

  • exact: keep output unchanged
  • compact: allow Alma to shrink large text fields
  • passthrough: skip compaction entirely

Preferred method:

json
{
  "name": "read_repo_file",
  "_meta": {
    "alma/outputSafetyMode": "exact"
  }
}

Fallback method:

json
{
  "name": "read_repo_file",
  "description": "Read repository file contents exactly [alma-output-safety: exact]"
}

Use exact for canonical source text, IDs, and structured outputs that later steps depend on exactly. Use compact for logs, web content, search results, and other lossy-safe text.

Tips ​

  • Only install servers from trusted sources
  • Review server permissions before enabling
  • Disable servers you're not actively using
  • Keep servers updated for security and new features