Most issues can be resolved using the solutions below. If you need more help, check the health check, browse the FAQ, or contact our support team at support@atatus.com.

Common issues

Symptom Likely cause Fix
Client says "no Atatus tools available" Incorrect configuration or the AI client was not restarted Verify your configuration file and perform a full restart (quit and reopen) of your AI client.
Invalid API key on every call Key is incorrect, expired, or for a different account region Generate a new Application Key in Settings » Account Settings » API Keys. Make sure the region matches your account.
Not Acceptable from a raw curl test The request did not advertise that it accepts the Server-Sent Events (SSE) stream Include -H "Accept: application/json, text/event-stream" in your curl request. Standard MCP clients handle this automatically.
The assistant says it found nothing The time range is too narrow, or telemetry has not been collected yet Ask the assistant to search a wider time window (like the last 7 days) and confirm that Atatus is receiving telemetry.
A high-level summary comes back empty Overall summary views can be sparse on some accounts Ask for specific details instead (e.g., "list the hosts" or "show unhealthy workloads").
A request takes more than 30 seconds The requested time range or volume of results is too large Narrow your query to a smaller time window, such as the last hour.
Replies stop with "Rate Limited: too many requests" Too many requests were sent in a short period Wait a moment before retrying. Reduce query frequency and use narrower time ranges.
Logs for a specific pod come back empty Pod name is not a standard log field Search by service or host name instead, as these are indexed reliably.

Health check

Confirm the server is reachable:

copy
icon/buttons/copy 
curl https://mcp.atatus.com/health

A healthy server returns {"status":"ok",...}.

Inspect the raw tool listing

To verify that the server is exposing all 44 tools correctly, run the following test in your terminal:

copy
icon/buttons/copy 
curl -s -X POST https://mcp.atatus.com/mcp \
  -H "X-API-KEY: ak_live_xxxxxxxxxxxxxxxxxxxx" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

If successful, this command returns the list of tools. Note that the response may be returned as an event: or data: Server-Sent Events (SSE) stream, which is normal behavior. If you receive an Invalid API key message, check that your key is correct and matches your account's region.

After changing your client config

Always restart your AI client completely after modifying its MCP configuration (such as mcp.json). Be sure to perform a full quit of the application rather than just closing the window.

Frequently asked questions

What is the Atatus MCP server?

The Atatus MCP server links your Atatus observability data to AI assistants that support the Model Context Protocol. Once connected, you can ask questions in plain English, and your assistant will query your APM, infrastructure, Kubernetes, logs, and traces to provide answers. For a high-level look, check out our Overview.

Which AI clients are supported?

Any client that supports remote MCP over HTTP will work. This includes VS Code, Cursor, Antigravity, Claude Code, Antigravity CLI, Codex CLI, and Claude Desktop. You can find configuration details for each of these in our Installation guide.

Which key should I use, and how does it differ from a License Key?

You must use an Application Key, which is specifically designed for querying data. You can manage these under Settings » Account Settings » API Keys. Do not use a License Key, as those are reserved for agents to send telemetry into Atatus.

Is the connection read-only?

Yes, the server is read-only by default. A Read key allows the assistant to query your data without making any changes. Currently, the only supported write action is resolving or ignoring errors, which requires a Read & Write key. We recommend using a Read key unless you want the assistant to perform these updates.

What data does the assistant see?

The assistant only receives the specific APM, infrastructure, Kubernetes, log, and trace data needed to answer your query. The server only returns data owned by your specific account. The subsequent use of this data by your AI client is governed by that client's privacy and data-handling terms.

How many tools are available, and what can they do?

The server provides 44 tools across APM, Kubernetes, Infrastructure, Logs, Distributed Tracing, and AI-powered analysis. You can review the full list of tools and sample questions in our Tool Reference.

Is my API key secure?

Yes, your key is transmitted securely over HTTPS in the X-API-KEY header. It only accesses your account's data, and you can revoke it instantly under Settings » Account Settings » API Keys.

Does it work on a non-US region?

Yes, but you will need to use your region's specific MCP endpoint. If you are unsure which endpoint applies to your account, please reach out to us at support@atatus.com.

Can I call the server directly using curl?

Yes. However, because the server uses Streamable HTTP, your request must include the header Accept: application/json, text/event-stream. While MCP clients send this header automatically, you must specify it manually when using curl. For more details, see the Inspect the raw tool listing section.

My client still cannot connect. Is the server enabled for my account?

The hosted MCP server is located at https://mcp.atatus.com/mcp. If you have configured your client correctly but still cannot connect, contact support@atatus.com to confirm that hosted MCP is active for your account.