MCP server

The FuseScope MCP server connects AI assistants to FuseScope over the Model Context Protocol (MCP). An assistant such as Claude Code, Claude Desktop, Cursor or VS Code can then answer questions from your telemetry and, if you allow it, change alerts, dashboards and channels for you.

Use this page when you want to ask an assistant about your services in plain language, or let it draft dashboards and alert rules.

What you can ask

• “Which services had the most errors in the last hour?”
• “Show me error logs for payment from the last 30 minutes.”
• “Get the trace for request 4bf92f3577b34da6a3ce929d0e0e4736 and tell me where the time went.”
• “What is the p99 latency of checkout today compared with yesterday?”
• “Which metrics drive most of our ingestion volume, and are they used in any dashboard?”
• “Create a dashboard with CPU and memory per host.”
• “How do I send Docker logs to FuseScope?”

The assistant picks the tools it needs: it searches logs and traces, runs metric queries, reads alert state and history, and looks up these docs.

How it works

The server is a single binary, fusescope-mcp-server. It calls the same FuseScope API as the UI, authenticated with a service account API key, so it can do only what that key’s role allows.

Transport	Runs	Use it when
stdio (default)	On your machine, started by the MCP client	One person, one FuseScope instance.
HTTP	As a shared service, /mcp endpoint	A team shares one server; optional OAuth so each user signs in with their own key.

Setup for both is in Set up the MCP server.

Read-only or read-write

Tools are marked read-only, create, update or delete with standard MCP tool annotations, so clients can auto-approve reads and ask before writes. What a write can actually do is decided by the role of the API key:

Role	What the assistant can do
Viewer	Query telemetry and read dashboards, alert rules, saved views and channels. Every write tool fails with PERMISSION_DENIED.
Editor	Also create, change and delete dashboards, saved views and alert rules.
Admin	Also manage notification channels.

Tip

Start with a dedicated service account with the Viewer role. Grant Editor only to the assistants you want to change things.

Privacy

• The server sends no product analytics or usage data anywhere.
• It exports its own traces and metrics only when you set an OTLP endpoint, for example to your FuseScope.
• Besides your FuseScope API, it fetches these docs (to answer how-to questions) and, when asked to import a dashboard template, the template file from GitHub. See Security to turn the docs refresh off.
• Telemetry the assistant reads is sent to the assistant’s model provider like any other conversation content. Choose the role and the data accordingly.

The server is open source under the Apache License 2.0: fuseone-io/fusescope-mcp-server.