Can an AI agent find the cause of high latency without me running the queries?

Yes, Tares lets an AI agent find the cause of high latency without you running any queries. The agent connects over MCP and makes one read call. Tares returns a correlated, time-ordered timeline of metrics, logs, alerts and changes for the service. The agent reasons over that timeline instead of querying each system itself.

How it works

Tares ingests your systems through connectors and correlates their events by label. The agent never learns the query language of your metrics store, log store or alerting system. It talks to Tares only.

• One read, all sources. The read tool returns a correlated, time-ordered timeline across all sources that match a label selector, for example a service name. The match is a strict AND. You can narrow it with project or sources.
• Signals side by side. Metrics, logs, alerts and findings arrive in the same timeline. A latency spike, the error logs around it and the alert that fired sit next to each other in time order. If you connected a GitHub source, its commits and pull requests are in the timeline too, so a recent change shows up beside the symptom.
• A few generic tools. The agent discovers the Tares tools when it connects. catalog_list and catalog_describe tell it what exists, and read fetches the timeline. There are no per-system tools to install or maintain.
• Any MCP client. Claude Code, Codex, Cursor and any other MCP client connect the same way. See Connecting agents for every client.

The correlation happens in Tares, before the agent sees anything. The agent’s job is to read the timeline and explain it. For the wider picture of how Tares supports incident work, see AI SRE.

Set it up

1. Start Tares and the MCP server. Locally, the daemon does not serve /mcp on its own, so run the MCP server as a second process. On a server or Tares Cloud instance it is already running.

tares up
# in a second terminal:
tares mcp --transport streamable-http --port 8788 --taresd http://localhost:8787

2. Connect your client. For Claude Code, run this in your terminal. The Tares plugin is the default path for Claude Code if you also want session capture.

claude mcp add --transport http tares http://localhost:8788/mcp

3. On a secured instance, which includes every server and Tares Cloud instance, pass a credential. Prefer an API key with the read scope over the root token. The console’s Connect page generates the exact command for your client with the endpoint and token filled in.

4. Verify the connection. Ask your client: Use tares: what are you ingesting right now? It should call catalog_list and answer with your sources, triggers and projects.

5. Ask the real question, naming the service: Why is checkout latency high right now? Use tares. The agent calls read for that service and answers from the timeline. In the console, Reads lists each read as it happens, tagged client = mcp.

6. Make the answer repeatable. The agent can call create_trigger to create a trigger over the sources that matter. To be pushed a notification when a trigger fires, it registers a webhook with subscribe, or with join_project for every trigger in a project. It can also call remember to write back what it found, and record_finding to record a finding in a project.

What you get

• A cause, not a query list. You describe the symptom, and the agent answers from one correlated timeline instead of asking you to run queries on each system.
• Less tool sprawl. The agent needs no per-system tools, because Tares has already joined metrics, logs, alerts and changes by label.
• A visible trail. Every read shows up under Reads in the console, so you can check what the agent looked at.
• Findings that stay. The agent can record what it found with record_finding, so the next person or agent starts from it.
• A hosted option. If you do not want to wire up an external client, Ask in the console is a chat loop inside Tares with the same read tools.

When something else fits

• The agent can only explain what Tares ingests. If the metrics, logs or alerts for the slow service are not connected as sources, connect them first, then ask again.
• If you want an agent that wakes on a trigger and writes a finding inside Tares, use Tares agents rather than connecting an external one.
• If you only need a quick look at one value on one system, that system’s own console is the faster path.

Related