Orion Agent Kit
MCP server providing Celestia network intelligence, including daily briefs, anomaly signals, staking and DA metrics, and live namespace and validator drill-downs, with each figure linked to its source.
README
<div align="center">
Orion Agent Kit
Celestia network intelligence for AI agents. An MCP server over Orion's daily analysis: anomaly signals, network verdicts, staking & DA metrics, live namespace and validator drill-downs — each figure linked back to the source it came from.
https://orionlabsone.cc/mcp
</div>
What this is
Orion watches Celestia mainnet continuously and publishes a daily machine-written brief: what moved, what drifted from baseline, and why it matters. This kit exposes that analysis to AI agents over the Model Context Protocol.
Think of it as an analysis layer for the Celestia agent ecosystem. It sits alongside knowledge bases (which answer "what is CIP-21?") and raw-data or transaction kits (which list blobs or build a delegate tx). Orion answers a different question: what is happening on the network right now, is it normal, and how did we get here?
Every number the server returns carries a provenance link to the Celenium or Celestia LCD endpoint it was read from, so the agent can cite its sources — and so can you. The server is read-only: no wallets, keys, or transactions anywhere in it.
Tools
Today & anomalies
| Tool | Question it answers |
|---|---|
get_daily_brief |
What's happening on Celestia today? (set include_narrative for Orion's written prose) |
get_signals |
Any anomalies (info / notable / critical) today or on date X? |
get_network_state |
Current height, DA volume, validators, staking ratios, Nakamoto coefficient, TIA price |
Trends & history
| Tool | Question it answers |
|---|---|
get_trends |
What changed since yesterday / last week? (day- and week-over-week deltas) |
get_metric_history |
Chart one metric (fees, blobs, price, staking ratio…) over N days |
list_reports |
What daily reports exist in the archive? |
get_report |
Full report for a given date |
search_reports |
When did the archive last mention "eclipse" or "block time"? |
Live drill-down (straight from Celenium)
| Tool | Question it answers |
|---|---|
get_top_namespaces |
Which rollups post the most data to Celestia right now? |
get_namespace |
Detail for one namespace: size, blob count, last activity |
get_top_validators |
Largest validators by stake — moniker, commission, jailed status |
Resources & prompts
Beyond tools, the server exposes MCP resources an agent can read or subscribe to:
orion://latest— the most recent daily reportorion://reports— the archive indexorion://report/{date}— any archived report by date
…and ready-made prompts to guide an agent:
daily_celestia_briefing— produce a grounded, source-cited daily briefingrollup_da_check— survey who dominates blobspace and the current DA cost context
What counts as a signal
Orion's detectors are event-driven, not state-driven — a persistently dominant namespace or a steady utilization day produces no signal, so what you see is what actually changed:
- DA utilization & fees — per-block volume and fees over the last day vs the recent daily baseline
- Blob-volume z-score — statistical spikes/drops in data posted
- Namespace shift — only when the DA leader changes or its share moves materially
- Staking & concentration — bonded-ratio moves, Nakamoto-coefficient changes, new jailings
- Market & protocol — notable TIA price moves, active governance proposals, new celestia-app releases
Each daily brief is written by an LLM narrator and passes an automated citation check — every number in the prose is verified against the structured data before publishing.
Quick start — hosted endpoint (no install)
The MCP server runs at https://orionlabsone.cc/mcp.
Claude Code
claude mcp add --transport http orion https://orionlabsone.cc/mcp
Cursor — add to .cursor/mcp.json (already included if you cloned this repo):
{ "mcpServers": { "orion": { "url": "https://orionlabsone.cc/mcp" } } }
VS Code — add to .vscode/mcp.json:
{ "servers": { "orion": { "type": "http", "url": "https://orionlabsone.cc/mcp" } } }
Claude Desktop / Codex / Windsurf / Cline / any stdio-only client — bridge with mcp-remote:
{
"mcpServers": {
"orion": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://orionlabsone.cc/mcp"]
}
}
}
Per-client walkthroughs: Claude Code · Claude Desktop · Cursor · Codex · VS Code
Quick start — clone & chat
git clone https://github.com/orioninfra/orion-agent-kit
cd orion-agent-kit
npm install
claude # Claude Code picks up .mcp.json automatically
Then just ask:
"Any anomalies on Celestia today?" · "What changed on Celestia since yesterday?" · "Which rollups post the most data right now?" · "How concentrated is the validator set?" · "When did the archive last flag block time?"
The bundled /brief and /signals commands wrap the two most common queries.
Run the server yourself
npm install
npm run start # Streamable HTTP on 127.0.0.1:8791/mcp
npm run start:stdio # stdio transport (for local MCP clients)
npm run smoke # end-to-end check of the tools against the live API
npm run typecheck # type-check the sources
Environment: PORT (default 8791), HOST (default 127.0.0.1), ORION_API_BASE (default https://orionlabsone.cc), CELENIUM_BASE (default https://api-mainnet.celenium.io/v1, used by the drill-down tools).
REST, without MCP
The underlying API is plain JSON, usable directly or via function calling with the OpenAPI spec:
| Endpoint | Returns |
|---|---|
GET /api/latest |
today's full report |
GET /api/reports |
archive index |
GET /api/reports/{date} |
full report for a date |
GET /api/rss |
the brief as an RSS feed |
How it fits together
AI agent (Claude / GPT / any MCP client)
│ MCP — read-only analysis tools, resources & prompts
▼
orion-agent-kit ──► Orion daily analysis (orionlabsone.cc/api)
│ │ every figure linked to its source
│ ▼
│ Celestia LCD · market data
└──────────► Celenium API (live namespace & validator drill-down)
License
Apache-2.0, Copyright 2026 Orion Labs. Questions, ideas, or a tool you wish existed: open an issue.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。