MCP Capabilities

GitLab MCP Server implements 6 MCP protocol capabilities that enhance how AI assistants interact with GitLab, plus icon metadata on every tool, resource, and prompt. These go beyond basic tool calling to provide richer, more intelligent interactions.

Capabilities overview

Capability	Direction	What It Enables
Logging	Server → Client	Structured log messages sent to the MCP client for visibility
Progress	Server → Client	Real-time progress updates for long-running operations
Roots	Client → Server	Workspace context — auto-detect GitLab project from local git repo
Sampling	Server → Client	AI-powered analysis — server sends GitLab data to client’s LLM for analysis
Elicitation	Server → Client	Interactive wizards — step-by-step forms for creating complex resources
Completions	Client → Server	Argument autocompletion for project names, branches, users, and more
Icons (metadata)	Server → Client	SVG icons for every tool, resource, and prompt

How capabilities work

Capabilities are negotiated during the MCP initialization handshake between the client and server:

sequenceDiagram
    participant C as MCP Client
    participant S as GitLab MCP Server

    C->>S: initialize (client capabilities)
    S-->>C: initialize response (server capabilities)
    Note over C,S: Both sides know what the<br/>other supports

    C->>S: Tool call
    S-->>C: Progress notification
    S-->>C: Log message
    S-->>C: Tool result

Server-declared capabilities (Logging, Completions) are always available. Client-dependent capabilities (Roots, Sampling, Elicitation) require the MCP client to declare support — the server checks for their presence before using them and gracefully degrades when they are unavailable.

Client support

Not all MCP clients support every capability. The server adapts automatically:

Capability	Claude Desktop	VS Code Copilot	Cursor	Claude Code
Logging	✅	✅	✅	✅
Progress	✅	✅	✅	✅
Completions	✅	✅	❓	✅
Roots	✅	✅	❓	✅
Sampling	✅	❌	❌	✅
Elicitation	✅	❌	❌	✅
Icons	✅	✅	✅	✅