How can my coding agent get a workspace overview in Rius?

Your coding agent gets a workspace overview in Rius by calling the workspace_metrics_overview tool on the Rius MCP server. Ask something like “How is my workspace doing today?” and the client picks the tool and returns a Markdown summary of volume, cost, tokens, latency and error rates. Each figure comes with the previous window for comparison. You can ask about the last 24 hours, a week, or the last 10 minutes.

How it works

The Rius MCP server exposes workspace_metrics_overview, which returns the aggregated health of one workspace over a time window. You rarely call it by hand. Your client reads your question, chooses the tool and fills in the arguments. The reply is Markdown, which your client renders as tables and summaries.

The main parameters are:

• workspace: a workspace name or ID. Leave it out to use your default workspace. If the same name exists in two of your organizations, pass org/workspace.
• hours: how far back to look, ending now. The default is 24, and “this week” is hours: 168.
• minutes: a lookback in minutes instead of hours, for short windows such as the last 10 minutes. It is ignored when 0.
• breakdowns: includes the ranked tables. The default is true.
• series: adds per-bucket sparkline values. The default is false, because they are verbose.

The server picks the bucket size from the window length: minute buckets up to 2 hours, hour buckets up to 48 hours, day buckets beyond that. Day buckets are UTC calendar days, so the first and last cover only part of a day, and a low value there is that clip, not a drop.

Sign-in is your Rius account over OAuth, or a read-scoped API key for agents that run without a browser. A key reaches the one workspace it was created in. The full list of tools and arguments is in the tool reference.

Set it up

1. Connect your agent to the Rius MCP server. If you already use the Rius plugin for Claude Code, Codex (beta) or Cursor (beta), it registers the server as rius and you can skip this step. For Claude Code without the plugin, run:

   claude mcp add --transport http rius https://mcp.eu.console.rius-glassflow.com/mcp

   For Codex, run codex mcp add rius --url https://mcp.eu.console.rius-glassflow.com/mcp. Other clients, and the headless key setup, are on the MCP page.

2. Check the connection by asking “Who am I in Rius, and which workspaces can I access?” Your client calls get_me and answers with a workspace table.

3. Ask for the overview in plain words. These questions work:

   • “How is my workspace doing today?”
   • “Anything wrong in the last 10 minutes?”
   • “How did my workspace do this week?” (your client sends hours: 168)

4. If a question does not pick the tool you expect, name it. For “How is my workspace doing today?” the client sends this call, with no arguments, so the defaults apply:

   { "name": "workspace_metrics_overview", "arguments": {} }

   For a short window, it sends the minutes argument instead:

   { "name": "workspace_metrics_overview", "arguments": { "minutes": 10 } }

5. Read the reply, then drill down with list_agent_traces and get_agent_trace, as the next section describes.

What you get

The reply is the set of numbers behind the dashboard’s overview, in one Markdown answer:

• Volume: traces, spans and LLM calls.
• Cost and tokens. Cost is a list-price estimate in US dollars, not a billed amount.
• Latency and time to first token: percentiles, in milliseconds.
• Error rates.
• A baseline for each figure: the previous window’s value and the change from it, so your agent can tell whether today is worse than yesterday.
• Ranked breakdowns (unless you turn them off): per model, top agents, calls by kind, top exceptions and slowest spans.

Two reading rules keep your agent honest. A dash means a figure was not measured, which is never the same as a measured 0. A metric listed as degraded failed to compute and is absent rather than zero.

The overview is the first step of an investigation. Once it shows a spike in errors or cost, your agent can list the failing traces with list_agent_traces and open one with get_agent_trace. See Monitor AI agents in production for how that fits into a production routine.

When something else fits

• To measure one agent, one service or one instance, use agent_traces_summary. It takes agent, service and instance filters and can return figures per time bucket.
• To see what one end user experienced, use user_activity_summary.
• To look at individual failed runs rather than workspace totals, use list_agent_traces and then get_agent_trace.

Related