How to Connect an Event-Intelligence MCP Server to Claude, ChatGPT, Cursor, and the Command Line

An AI agent's knowledge stops at its training cutoff, so it cannot tell you what happened in the world this morning. The Off-Nadir Delta MCP server fixes that by giving the agent live, geolocated world-event intelligence on demand — and it is now reachable through four different connection methods, from a one-click OAuth approval in a chat app to a single command line. This guide shows each one and when to use it.

What is the Off-Nadir Delta MCP server?

It is a Model Context Protocol (MCP) server that exposes live world-event intelligence to AI agents as callable tools. Over one endpoint an agent can query geolocated event signals, find where activity is concentrating, search satellite imagery over an event, read the AI-synthesized Daily World Brief, and ask an OSINT/GEOINT analyst — grounded in sourced, geolocated signals rather than a static training snapshot. The Model Context Protocol is an open standard for connecting AI applications to external tools and data; see the MCP specification for the protocol itself.

The server ships 8 tools, 5 resources, and 4 prompts. Every tool is available on every plan, including Free — the only gate is your token balance, and two of the tools cost nothing to call.

Which connection method should you use?

Pick the method that matches how you run your agent. All four reach the same intelligence and the same token balance underneath.

Where your agent runs	Method	Auth	Best for
Claude app / ChatGPT	Remote connector (OAuth 2.1)	Approve in-app, no key to copy	Chatting with situational awareness
Cursor / VS Code	One-click install deeplink	OAuth or bearer key on first use	Coding agents in an editor
Claude Code / any CLI	claude mcp add (Streamable HTTP)	Bearer API key	Terminal and scripted agents
Local / self-hosted	npx or Docker (stdio)	Bearer API key via env	Clients that prefer a local process

How do you add it to Claude or ChatGPT?

Add it as a remote connector — no API key to paste. In claude.ai or ChatGPT (Developer Mode), open Settings → Connectors → Add custom connector and enter the endpoint:

https://offnadir-delta.com/api/v1/mcp

The chat app then walks you through an OAuth approval and the tools appear in your conversation. The server implements OAuth 2.1 with PKCE and Dynamic Client Registration, the authorization pattern the MCP authorization specification defines for interactive clients, so the app registers itself and you simply approve access. Custom connectors are available on Claude's and ChatGPT's paid plans; the connection is metered on your token balance and revocable anytime. Step-by-step screens are in the MCP setup docs.

How do you install it in Cursor or VS Code?

Use the one-click install buttons on the MCP overview page. Cursor installs the server from an anysphere.cursor-deeplink install link; VS Code installs it from a vscode:mcp/install link with the HTTP transport pre-filled. You complete OAuth — or paste your ond_… API key — on first use. This is the fastest path for a coding agent that already lives in your editor.

How do you connect it from the command line?

For Claude Code or any MCP client that speaks HTTP, add the server in one line with your API key as a bearer header:

claude mcp add --transport http off-nadir-delta \
  https://offnadir-delta.com/api/v1/mcp \
  --header "Authorization: Bearer ond_..."

Issue a key first from Account → Developer API; keys are shown once and revocable anytime. Any client that supports HTTP transports works the same way — point it at /api/v1/mcp and pass the Authorization header.

How do you run it locally with npx or Docker?

If your client prefers a local stdio server over a remote connector, run the published package. It is a thin local server that forwards calls to the hosted endpoint with your key:

{
  "mcpServers": {
    "off-nadir-delta": {
      "command": "npx",
      "args": ["-y", "offnadir-delta-mcp"],
      "env": { "OFFNADIR_DELTA_API_KEY": "ond_your_key_here" }
    }
  }
}

The package is on npm as offnadir-delta-mcp, and a Dockerfile is included for containerized setups (docker run --rm -i -e OFFNADIR_DELTA_API_KEY offnadir-delta-mcp). Tool discovery works without credentials, so a registry or client can inspect the server before you add a key; only metered tool calls need the key.

Is there a Python SDK?

Yes. The official Python client is on PyPI as offnadir-delta and installs with pip install offnadir-delta. It queries the same event signals, hotspots, statistics, and satellite imagery from Python with both synchronous and asynchronous interfaces — useful when you are wiring the intelligence into your own code rather than an agent. The server is also listed in the official MCP Registry as com.offnadir-delta/mcp, which is the machine-readable index that MCP clients and directories read from.

Which tools, resources, and prompts are available?

The server exposes eight tools. Two are free to call, so an agent can orient itself before spending anything:

Tool	Cost	What it does
get_world_brief	Free (0 tokens)	Read the AI-synthesized Daily World Brief
get_usage	Free (0 tokens)	Check remaining token balance and plan capabilities
query_signals	Metered	Geolocated event signals by area, date, and category
query_stats	Metered	Event counts by category and day
query_hotspots	Metered	Where activity is concentrating, grid-binned and ranked
search_imagery	Metered	Satellite scenes (Sentinel-1/2, OPERA RTC-S1) over an area
assess_signal	Metered	AI remote-sensing assessment for one event
ask_analyst	Metered	Agentic OSINT/GEOINT Q&A returning a structured, cited brief

It also ships resources the agent can read directly — brief://latest, the templated brief://{date}, signals://schema, usage://current, and imagery://collections — and prompts that orchestrate the tools for common workflows: daily-situation-briefing, assess-top-signal, aoi-watch, and market-exposure-check. For the tool ergonomics behind these, see How to Give an AI Agent Live World-Event Awareness.

How do you verify it is working?

Start with a free call so it costs nothing. Once the server is connected, ask your agent to read the latest Daily World Brief — that invokes get_world_brief, which draws zero tokens. A good first prompt: "Using the Off-Nadir Delta connector, read the latest Daily World Brief, then list the five highest-severity world-event signals from the last 24 hours, each with its coordinates, category, and a source link." The first step is free; the second (query_signals) draws from your balance. Check your balance any time with get_usage, also free. If you are new to the underlying concepts, the glossary defines the event-intelligence terms these tools return.

Where to go next

• Connect it now: the MCP setup docs have the per-app screens and config blocks.
• See the tool surface: the MCP overview lists every tool, resource, and prompt with copyable demo prompts.
• Build a workflow: From a World Event to the Satellite Scenes Over It chains the tools end to end.
• Try the analyst: Delta Agent is the in-app version of the same analysis the tools expose.

Off-Nadir Delta is an independent project by Kazushi Motomura. Reading the Daily World Brief and checking usage are free; queries, assessments, and analyst calls are metered on your token balance, with a Free tier included — so you stay in control of what an agent can spend.