Summary

Read-only unless you say otherwise, and it masks the columns that look like secrets.

A security gate blocks DROP, DELETE, INSERT and the rest before they reach the database, and columns named like passwords, tokens or card numbers come back masked — both on by default. describe_table is the tool to start with: it returns the schema together with sample rows and guidance, which is most of what you need to write a correct query against an unfamiliar time-series table.

What it is

A GreptimeDB server for metrics, logs and traces. It runs queries in three dialects, describes tables with the semantic metadata that time-series schemas carry, and manages ingest pipelines and Perses dashboards over the HTTP API.

What you get

execute_sql — SQL with a format option (csv, json or markdown) and a row limit
execute_tql — PromQL-compatible queries for time-series analysis
query_range — time-window aggregation with RANGE and ALIGN
describe_table — schema, semantic metadata, recent sample rows and query guidance in one call
explain_query — the execution plan for a SQL or TQL query, with runtime statistics and index-pruning counters when you ask for them
health_check — connection status and server version
list_pipelines, create_pipeline, dryrun_pipeline, delete_pipeline — manage ingest pipelines, and test one against sample data before it writes
list_dashboards, create_dashboard, delete_dashboard — Perses dashboard definitions
Prompt templates for common jobs, including log_pipeline, metrics_analysis, promql_analysis, trace_analysis and query_performance_tuning

Requirements

A reachable GreptimeDB instance. Install from PyPI as greptimedb-mcp-server and pass connection details as CLI arguments or environment variables — GREPTIMEDB_HOST, GREPTIMEDB_PORT (the MySQL protocol port), GREPTIMEDB_USER, GREPTIMEDB_PASSWORD, GREPTIMEDB_DATABASE. Pipeline and dashboard management also needs GREPTIMEDB_HTTP_PORT. Runs over stdio, SSE or streamable-http.

Setup effort

One command — pip install greptimedb-mcp-server