Orion Agent Kit MCP server

What it does

An MCP server that exposes Orion's daily analysis of the Celestia network, including anomaly detection, validator metrics, staking ratios, and data availability trends. Intended for AI agents that need to understand current network state and historical patterns on Celestia mainnet.

• Query today's network activity on Celestia and identify anomalies (info, notable, or critical severity) to surface unexpected behavior.
• Track metrics like DA fees, blob volume, staking ratios, and Nakamoto coefficient over days or weeks to spot trends.
• Drill into live namespace and validator data to see which rollups post the most data and which validators hold the largest stakes.

Tools this server exposes

Tool	Does
get_daily_brief	Returns today's daily analysis of Celestia network activity, with optional narrative prose from Orion's LLM writer.
get_signals	Lists anomalies detected on a given date, categorized by severity (info, notable, critical).
get_network_state	Returns current network metrics: block height, DA volume, validator count, staking ratios, Nakamoto coefficient, and TIA price.
get_trends	Shows day-over-day and week-over-week changes in network metrics.
get_metric_history	Charts a single metric (fees, blobs, price, staking ratio) over a specified number of days.
list_reports	Returns a list of available daily reports in the archive.
get_report	Retrieves the full report for a specific date.
search_reports	Searches the report archive for a keyword or phrase.
get_top_namespaces	Lists the rollups currently posting the most data to Celestia, ranked by activity.
get_namespace	Returns details for a single namespace: total size, blob count, and last activity time.
get_top_validators	Lists the largest validators by stake, including moniker, commission rate, and jailed status.

Before you enable it

An MCP server runs with whatever access you give it. Start with a development account rather than production, and check which tools the server exposes before you let an agent call them unattended.