Superlog
by superloglabs
ask your coding assistant why production broke.
Cloud and DevOpsOfficialRemote
1. Made by
“"url": "https://api.superlog.sh/mcp"”
The server code lives in the vendor's own repository (superloglabs/superlog, apps/api/src/mcp), and the registry entry sh.superlog/superlog points at api.superlog.sh, the vendor's domain.
github.com/superloglabs/superlog/blob/78123eebd23d41c54bd09c70a563ba311d7546e6/server.json2. Keys
“"url": "https://api.superlog.sh/mcp"”
The configs come from the install panel in Superlog's own web app code; the README itself has no MCP setup.
github.com/superloglabs/superlog/blob/78123eebd23d41c54bd09c70a563ba311d7546e6/apps/web/src/onboarding/McpInstallDialog.tsx3. Runs
Endpoint: https://api.superlog.sh/mcp. Checked: yes: an unauthenticated HEAD to https://api.superlog.sh/mcp answered HTTP 401 on 2026-11-10 (host up, sign-in required).
api.superlog.sh/mcp4. Commercial use
“Superlog is licensed under the [Apache License 2.0](./LICENSE.md).”
The community edition may be used, changed and run in commercial settings under Apache-2.0.
github.com/superloglabs/superlog/blob/78123eebd23d41c54bd09c70a563ba311d7546e6/README.mdChecked against its repository at one commit, on Nov 10, 2026. Tap a verdict to see its source.How we checkedReport an error
What Superlog does
Superlog collects the logs, traces and metrics your app sends while it runs and groups errors into incidents. Its MCP server lets a coding assistant such as Claude Code, Codex or Cursor search that data and explain an incident. With write access it can also add alerts and dashboards for you.
What people use it for
Find the cause of an error
Ask your assistant what failed in the last hour. It searches the logs and request traces of your app instead of you pasting logs into the chat.
Explain an incident from its link
Paste the link of an incident from Superlog and the assistant reads it, with what the investigation agent found.
Turn a fix into an alert
After a bug is fixed, the assistant adds an alert so the same failure warns you next time. It can try the alert on current data before saving it.
Quiet noisy errors
When the same harmless error keeps opening incidents, the assistant changes the filter that decides what counts as an issue. It shows the effect on the last day first.
A dashboard for one service
Ask for a dashboard of errors and slow requests for one part of your app. The assistant creates it and adds charts and tables.
Teach the investigator about your system
Write down how your system is built and what you learned from past bugs. Superlog's own investigation agent reads it on every later run.
Superlog popularity
Tools it gives your agent (48)
What it sends and its license
log lines, traces, metrics and incident details the tools return (may include user ids, emails or request data your app logged)
To your coding assistant and the company running its model; on every query tool call. Opt out: use a read-only or narrower token; keep personal data out of your logs.
an analytics event per tool call (user, org, project, tool name, success, time taken, error text)
To Superlog's product analytics (PostHog), when its token is set on the server; every tool call. Opt out: self-host without a PostHog token.
requests from Superlog to outside MCP servers an admin adds for the investigation agent
To those servers; add_agent_mcp_server, test_agent_mcp_server, start_agent_mcp_oauth, and later investigations. Opt out: do not add outside servers.
The community edition is open source under the Apache-2.0 license and free to run yourself, also in paid products. The hosted Superlog Cloud has a free tier and paid plans.
The community edition may be used, changed and run in commercial settings under Apache-2.0.
“Superlog is licensed under the [Apache License 2.0](./LICENSE.md).” source
Copies must carry the Apache-2.0 license and keep its notices.
“You must give any other recipients of the Work or Derivative Works a copy of this License” source
Superlog says it does not train foundation models on customer content, and neither may its subprocessors.
“We do not use Customer Content to train foundation models, and our subprocessors are not permitted to either.” source
Is it alive
1.0.0 in the registry, updated Jul 21, 2026
no GitHub releases published; last commit 2026-11-03 at the pinned sha
Alternatives
Superlog This page
io-github-grafana-mcp-grafana
Pick it if your telemetry already lives in Grafana
Pick it if you need live server and machine health, not app-level traces
Pick it if the app runs on Cloudflare and you need its own logs and settings
Get started
Tell your AI assistant
“Connect the Superlog MCP server (claude mcp add --transport http superlog https://api.superlog.sh/mcp), then search the last hour of error logs and traces for the checkout service and explain the most frequent error; do not change alerts or filters without asking me.”
Connect it
claude mcp add --transport http superlog https://api.superlog.sh/mcp{
"mcpServers": {
"superlog": {
"url": "https://api.superlog.sh/mcp"
}
}
}
{
"mcpServers": {
"superlog": {
"url": "https://api.superlog.sh/mcp"
}
}
}The configs come from the install panel in Superlog's own web app code; the README itself has no MCP setup.
Mistakes to avoid
The tool rejects other combinations.
Each bucket you send replaces the old one, and the server's own guidance says to preview first.
The tool replaces the whole field, so a partial text erases the rest.
HTTP status, route and method sit in span attributes, not in the span status.
Delete is permanent; archiving can be undone.
Questions
Does my assistant need a key?
No for Claude Code, Codex and Cursor, which sign in through the browser. A personal access token is the option for tools without that sign-in.
Can I run it on my own servers?
Yes. The community edition runs locally with Docker, Node.js 20+ and pnpm 9+, with its API on port 4100.
Do I have to drop Sentry or Datadog?
No. Superlog connects to the tools you already use and adds their context.
Is my data used to train AI models?
The privacy policy says Superlog does not use customer content to train foundation models, and its subprocessors may not either.