How can I check that Tares is really receiving the data an AI SRE needs for a disk filling up?
Tares lets you check ingestion yourself by reading the entity through an MCP client and looking at the timestamps of the events it returns. Recent timestamps from each source mean data is arriving. Old timestamps, or a source that is missing from the read, point to a source that stopped sending. You can cross-check with the source health and freshness tools in the same client.
How it works
The read tool returns a correlated, time-ordered timeline across all sources that match a label selector. Every event in it carries a timestamp, so one read shows when each source last delivered something about your entity.
- Newest event per source. For a disk filling up, find the sources that carry your disk signals, such as host metrics, logs or alerts. Each should show events from the last few minutes. A source whose newest event is old, or that does not appear at all, is not delivering.
- Ask for a recent window. Phrase the question with a time range, for example the last 15 minutes, so the timeline shows what is arriving now.
- Source health and freshness.
list_sourcesreturns each source with its config and live health.catalog_describereturns one object’s schema, entities, freshness, lineage and sample events. - Volume over time.
statscounts events per value of a label over a project’s sources, this window against the one before. A count that drops to zero against the previous window shows a source that went quiet. - Reads are visible. In the console, Reads lists each read as it happens, tagged
client = mcp, so you can confirm your check went through.
For how an AI SRE uses this data, see AI SRE.
Set it up
-
Start the MCP endpoint. On a local Tares, the daemon alone does not serve
/mcp, so run the MCP server as a second process aftertares up:tares mcp --transport streamable-http --port 8788 --taresd http://localhost:8787On your own server or on Tares Cloud, the MCP server is already running. Tares Cloud shows your endpoint in the console under Connect.
-
Connect your client. For Claude Code:
claude mcp add --transport http tares http://localhost:8788/mcpCursor, Codex and other MCP clients connect the same way, see Connecting agents (MCP). On a secured instance, give the client an API key with the
readscope. -
Confirm the connection. Ask your client: “Use tares: what are you ingesting right now?” It should call
catalog_listand answer with your sources, triggers and projects. Check that the sources for your disk signals are in that list. -
Read the entity. Ask for the entity that holds the disk signals, for example: “Use tares: what happened to my database host in the last 15 minutes? Show the timestamp of each event and name its source.” The client calls
readwith a label selector for that entity. -
Look at the timestamps. For each source you expect, check that the newest event is recent. If one source shows only old events, or nothing, that source stopped sending.
-
Cross-check. Ask the client to call
list_sourcesfor live health andcatalog_describeon the suspect source for its freshness and sample events. Ask forstatson a label such as the host to compare this window with the one before. -
Fix and repeat. Correct the source, then run the same read again and confirm new timestamps appear.
What you get
- A direct answer to whether data about your entity is arriving, taken from the timestamps in a real read.
- A way to tell which source went quiet, since each event in the timeline names where it came from.
- Health, freshness and sample events for any source, from the same client.
- A check you can repeat after every change to a source, without leaving your editor or terminal.
When something else fits
- If you would rather not wire up an external client, use Ask in the console, the in-app chat that has the read tools.
- For the full picture of how Tares watches for problems such as a disk filling up, start with AI SRE. The agents page lists every client and its connection command.
Related
- Connecting agents (MCP)
- Where do I see whether the trigger for a disk filling up fired and which run it woke?
- How do I label Grafana Loki logs so an AI SRE can narrow a disk filling up to one service?
- Which labels on Prometheus alerts about a disk filling up tie it to the right service?
Still stuck? Get help