ethpandaops Query Guide

This skill is a router. The full, always-current usage guide is owned by the code — read it first and follow it:

• CLI: panda getting-started
• MCP: read the panda://getting-started resource

It is generated live from the running server — workflow, discovery pointers, and sessions — so it never goes stale. Dataset query rules (table syntax, partition-key filters, FINAL, network filtering) live in per-dataset guides: panda datasets lists them, panda datasets <name> shows one. Use panda docs <module> for a module's full API and panda search examples "<topic>" for worked queries. Everything below is the durable context those guides do not carry.

Discover names — don't hardcode them

Datasource, cluster, and table names are owned by the proxy and change over time, so enumerate them from the live tooling rather than pasting a name from memory, an old chat, or a screenshot:

panda datasources                                 # datasources and their types
panda datasets                                    # datasets and where they live
panda datasets <name>                             # one dataset's query guide
panda schema [<cluster> [<database> [<table>]]]   # clusters → tables → schema

The embedded examples and docs (panda search examples, panda docs) are compiled from the current binary, so the names in their output are current too — trust those.

Search before writing queries

Working query patterns live in the embedded examples and runbooks:

panda search examples "block arrival time"
panda search runbooks "finality delay"

Logs are in ClickHouse, not Loki

Container logs from public devnets and platform services ship via OpenTelemetry into ClickHouse (external.otellogs) — there is no Loki datasource for them. For the schema and the full procedure: panda read runbooks://debugethereum_network. (Local Kurtosis devnet logs are the separate autodiscovered local-kurtosis datasource.)

Public-devnet xatu event data also lives on clickhouse-raw: each public devnet has its own database named after the network — backtick-quote it, e.g. ` FROM blob-devnet-0.beaconapiethv1events_block `.

Notes

• Prefer the CLI (panda binary); use the MCP tools (execute_python,

manage_session, search) only if they appear in your tool list.

• Each execution is a fresh Python process — variables do not persist, but /workspace/

files do. Default timeout 60s, max 600s.

• NEVER recite or paste base64 image data. Save the image to /workspace/ and

storage.upload() it to return it to the user.