Use OpenObserve from Claude Code: query logs, metrics and traces, and work with streams, dashboards and alerts through conversation.
| Skill | What it does |
|---|---|
setup |
Connects Claude Code to the MCP server built into your OpenObserve instance. Asks for your cloud region or self-hosted URL and your organization, checks the endpoint, and registers it. |
sql-query |
Schema-first workflow for writing and validating SQL and PromQL against your streams. |
dashboard-creation |
Builds and edits dashboards: layout grid, chart types, panel JSON, and the panel-level tools. |
alert-creation |
Creates and edits alerts and alert destinations, with sensible defaults. |
pipeline-creation |
Designs data pipelines: source and destination streams, functions and conditions. |
vrl-scripts |
Writes and validates VRL functions for parsing and transforming data. |
spl-conversion |
Converts Splunk SPL searches to OpenObserve SQL. |
openobserve-instrument |
Instruments an LLM or AI-agent app to send OpenTelemetry traces to OpenObserve. |
The skills are adapted from the @openobserve/skills package (version 1.2.1) for use in Claude Code.
/plugin marketplace add openobserve/claude-code-plugin
/plugin install openobserve@openobserve
In Claude Code, run:
/openobserve:setup
or just ask "connect OpenObserve". You will be asked for:
| Value | Example | Notes |
|---|---|---|
| OpenObserve URL | https://cloud.openobserve.ai |
Cloud: cloud (US), eu1, ap1. Self-hosted: your own host. |
| Organization ID | default |
The org_identifier shown in the OpenObserve UI address bar. |
Then run /mcp, pick openobserve, and sign in. Sign-in uses OAuth in your browser; the plugin stores no credentials.
To do the same by hand:
claude mcp add --transport http --scope user openobserve https://<host>/api/<org_id>/mcp
setuprunsclaude mcp list, one unauthenticatedcurlprobe of the endpoint you name, andclaude mcp add.openobserve-instrumentedits your project, installs OpenTelemetry packages, and runsscripts/preflight.py(Python standard library only), which sends an empty trace batch to your OpenObserve instance to check the credentials in your.env. For supported Python SDKs it downloads the OpenObserve installer from a pinned commit ofopenobserve/o2-datasourceand runs it only if its SHA-256 matches the value in the skill.- The other skills only call OpenObserve MCP tools.
Every OpenObserve organization has its own MCP address (<host>/api/<org_id>/mcp), and self-hosted instances live on their own hosts, so there is no single URL a plugin could ship. The setup skill registers your endpoint as an ordinary user-level MCP server, which you can inspect or remove with claude mcp list and claude mcp remove openobserve.
The MCP server acts with the permissions of the signed-in user. It is not read-only: besides search, it exposes tools that create, update and delete dashboards, alerts and other resources. Claude Code asks before each tool call unless you allow it; use a role with read-only access if you want to rule out changes.
claude plugin validate .
claude --plugin-dir .
The plugin talks only to the OpenObserve instance you configure. See the OpenObserve privacy policy.
Apache-2.0