doc/operations/observability/mcp_server.md
{{< details >}}
{{< /details >}}
{{< history >}}
{{< /history >}}
GitLab Observability provides a Model Context Protocol (MCP) server so you can query your observability data from AI assistants and agents using natural language.
After you connect, an MCP-capable client can ask questions about your telemetry and get answers backed by your own data. Supported clients include Cursor, Claude (Claude Code and Claude Desktop), VS Code, and Codex. For example:
The MCP server can query metrics, traces, logs, alerts, dashboards, and services in your GitLab Observability instance.
The MCP server has no standing credentials of its own. Each user authenticates with their own GitLab Observability API key, which their MCP client sends with every request. Access is scoped to what that key is allowed to see.
To create an API key:
[!warning] Store your API key securely. Do not commit it to version control. Use your MCP client's secret management or environment variable support if available.
Your MCP endpoint follows this pattern:
https://<namespace_id>.mcp.gitlab-o11y.com/mcp
Replace <namespace_id> with the ID of the namespace where you enabled
Observability. The namespace ID is your group ID, or your personal namespace ID
if you enabled Observability on a personal project.
SIGNOZ-API-KEY header.The MCP server exposes tools for reading and exploring your observability data, including:
Your AI assistant chooses the appropriate tools to answer your question and returns results drawn from your instance's data.