How can my coding agent list the versions of an agent in Rius?
Your coding agent lists an agent’s versions in Rius with the agent_versions tool on the Rius MCP server, one call per agent. Each version comes back with its traces, failed traces, failed rate and when it was first and last seen. You ask in plain words, and your client fills in the arguments.
How it works
The Rius MCP server exposes agent_versions. It returns the versions one agent ran in a time window, on two separate dimensions that are not synonyms:
- agent is the agent definition (prompt, tools, policy), as
rius.main_agent.versionstamps it. - service is the build that ran it, as
service.versionstamps it.
A prompt edit can ship without a deploy, and a deploy can ship without touching the agent, so the two move independently. Rius reports them in separate lists.
The main parameters:
agent(required): the exact agent name. Versions are per agent. Read the name fromlist_agentsor from theroot_agentcolumn oflist_agent_tracesinstead of guessing it.hours: the lookback window ending now, 24 by default. “This week” is 168.startandend: optional absolute RFC3339 bounds that override the hours-derived value.instance: optional exact instance id, to narrow to one running copy of the agent.workspace: optional workspace name or ID. Omit it to use your default workspace.
The tool is read-only. It is the same read the Agent page in the console draws its Versions figure from. For more on the tool surface, see the tool reference.
Set it up
-
Connect your coding agent to the Rius MCP server if it is not connected yet. In Claude Code, run:
claude mcp add --transport http rius https://mcp.eu.console.rius-glassflow.com/mcpIf you use the Rius plugin, it already registers this server as
rius, so skip this step. For a headless agent, use a read-scoped API key as a bearer header. If the plugin is also installed, give that key-based server its own name, such asrius-key. All other clients and the key setup are on the MCP page. -
Ask in plain words, naming the agent and the period:
- “Which versions of planner ran this week?”
- “When did planner’s new prompt version first show up?”
If you do not know the exact agent name, ask “Which of my agents are running right now?” first. Your client calls
list_agentsand you pick the name from the reply. -
Your client picks
agent_versionsand sends the arguments. For the first question it sends:{ "name": "agent_versions", "arguments": { "agent": "planner", "hours": 168 } }For the second, a longer window such as
"hours": 720reaches back 30 days, so an older version still shows up. -
Read the reply. It is Markdown, which your client renders as a table, and it lists the agent versions and the service versions separately. Pass the exact version strings it shows to
compare_agent_versions, asbaseandhead, if you want a side-by-side comparison.
What you get
Every row in the reply carries these figures for one version inside your window:
- traces
- failed traces
- failed rate
- when the version was first seen and last seen
Rows are ordered newest-seen first within each dimension and capped at 50 per dimension. The totals still count past the cap.
Every row is scoped to the root span of each run, so a row’s numbers equal agent_traces_summary filtered to that version. The figures your coding agent reports match what you read on the Agent page in the console.
You also get a quick answer to “when did this version first appear?” and the exact version values to hand to compare_agent_versions. To learn how Rius surfaces agents and versions across a whole system, see Monitor AI agents in production.
When something else fits
- To see how two versions differ on failed rate, cost per trace, latency and tokens, ask for
compare_agent_versions. It takes the agent, the dimension, and the base and head values thatagent_versionslists. - For one agent’s numbers over a window, or a single version’s numbers, use
agent_traces_summarywith a version filter. - To browse versions visually, open the Agent page in the console.