How can I confirm my MCP client is really connected and reading from Tares?

Tares confirms the connection when you ask your client what it is ingesting right now. The client calls catalog_list and answers with your sources, triggers and projects. Then the console Reads list shows each read tagged client = mcp. If both happen, your client is connected and reading live data.

How it works

Once your MCP client connects, it discovers the Tares tools. One of them, catalog_list, is a read tool that lists your sources, triggers and projects. A client that is really connected to Tares calls it when you ask what Tares is ingesting, and it answers from the result rather than from guesswork.

The check has two sides:

• In the client: the answer names your own sources, triggers and projects, and the client shows a catalog_list tool call.
• In the console: the Reads list records each read as it happens and tags it client = mcp. This shows the request reached Tares and was served.

Use the two together. An answer in the client confirms the client side. A tagged read in the console confirms Tares received it. For more on what the tools do and how clients connect, see Connecting agents (MCP).

Set it up

1. Make sure an endpoint is running. On a local Tares, the daemon alone does not serve /mcp, so start the MCP server as a second process:

   tares mcp --transport streamable-http --port 8788 --taresd http://localhost:8787

   On your own server or on Tares Cloud, the MCP server is already running and there is nothing to start. Tares Cloud shows your endpoint under Connect in the console.

2. Connect your client. For Claude Code over the CLI, run this against the local endpoint:

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

   Other clients, such as Codex, Cursor and Claude Desktop, have their own snippets on the agents page. The console’s Connect page generates the exact commands with your endpoint and token filled in.

3. Ask what it is ingesting. In your client, send:

   Use tares: what are you ingesting right now?

4. Check the answer. The client should call catalog_list and answer with your sources, triggers and projects. If the names match what you see in the console, the client is reading from your Tares.

5. Check the console. Open Reads. Each read shows up as it happens, tagged client = mcp. Find the read from the moment you asked.

6. Optionally, ask a real question. If you run the demo, ask “what happened to api-server in the last 15 minutes?” The client answers from the correlated timeline, which also shows up in Reads.

On a secured instance, the client must send a credential. Use an API key with the read scope as a Bearer token in the Authorization header. On an open local instance, no credential is needed.

What you get

• A clear yes or no on whether the client reaches Tares, from one question.
• Proof from both sides: the client’s answer lists your sources, triggers and projects, and the console records the read.
• A way to tell Tares reads from other activity, because each MCP read carries the tag client = mcp.
• A working base for the next step. Once the check passes, you can ask correlated questions about logs, metrics, alerts and findings in one read, as described in AI SRE.

When something else fits

• If the client answers without calling catalog_list, or the answer does not match your Tares, check the endpoint URL first. On a local Tares, confirm the second process is running. On a secured instance, confirm the credential is set.
• If you want to explore your data without wiring up an external client, use Ask in the console. It is a chat that runs inside Tares with the same read tools.

Related