MCP Server

Connect AI agents to
live news intelligence.

The NewsDesk MCP server gives any MCP-compatible client direct access to the structured news database — events, entities, relationships, actions, and graph queries. Ask questions in plain English; the agent handles the rest.

Overview

What is this?

The NewsDesk MCP server exposes the entire structured news database as callable tools. Any MCP-compatible client — Claude Code, Cursor, Windsurf, or a custom agent — can connect and query events, entities, relationships, and graph structures using natural language.

The agent automatically translates your questions into the right tool calls. You just ask questions in plain English.

Getting started

Setup in 2 minutes.

Tip You can simply paste this page's URL to your agent and ask it to set this up for you.
01
Create .mcp.json

In the directory where you run your MCP client, create a file called .mcp.json:

{ "mcpServers": { "newsdesk": { "type": "http", "url": "https://app.newsdesk.dev/mcp" } } }

No API keys or tokens to manage.

02
Restart your client

Claude Code, Cursor, and other MCP clients pick up .mcp.json on startup. Restart after creating the file.

03
Sign in

On first connection, your client will open a browser window for Google sign-in. Use the same Google account you use for the NewsDesk dashboard. Your session is cached — you won't need to sign in again until the token expires.

04
Start querying

The agent now has access to 17 NewsDesk tools. Try:

// Example prompts "What's happening with NVIDIA and China?" "How are Elon Musk and the SEC connected?" "Find all government agencies that took negative actions against tech companies" "Find the path between Donald Trump and Saudi Arabia"
Reference

What can it do?

17 tools across five categories. Most results include clickable links to the NewsDesk dashboard for visual exploration.

Category Capability
Entity lookup Search and inspect any of the 17 entity types — people, companies, countries, organizations, commodities, and more
Event discovery Semantic search across thousands of active events, find similar events, bulk fetch event details
Relationship exploration Browse entity connections, get full relationship details, measure co-mention strength
Action analysis Find who did what to whom — sanctions, acquisitions, agreements, appointments
Graph queries Declarative multi-hop pattern matching — "find all CEOs of companies connected to China"
Pathfinding Shortest paths between any two entities through the relationship graph
Usage

Example workflows.

Simple entity lookup
"Tell me about Tesla"

The agent searches for Tesla, retrieves its profile, and lists its recent events.

Relationship exploration
"How are the US and China connected?"

The agent finds both entities, retrieves their relationship data, mutual events, and actions between them.

Semantic search
"What events are related to AI regulation in Europe?"

The agent searches across all active events using natural language similarity.

Complex graph pattern
"Find all people who are CEOs of companies that have relationships with China"

The agent builds a multi-hop graph query and returns matching entities with dashboard links.

Pathfinding
"How is Elon Musk connected to the Federal Reserve?"

The agent finds shortest paths through the relationship graph and returns interactive visualizations.

Compatibility

Supported clients.

Any client that supports the MCP protocol with HTTP transport:

  • Claude Code — CLI, desktop app, web app, IDE extensions
  • Cursor
  • Windsurf
  • Custom agents built with the MCP SDK
Help

Troubleshooting.

Browser doesn't open for sign-in

Make sure your client supports OAuth for MCP. Some older versions may not — update to the latest version.

Tools not appearing

Check that .mcp.json is in the directory where you launched the client. Restart after creating or modifying the file. Verify that https://app.newsdesk.dev/mcp is reachable from your network.

"Entity not found" errors

Entity references are case-sensitive. Always let the agent search for entities first rather than guessing names. Some entities have unexpected names (e.g., Apple Inc. not Apple).

Slow first query

The first semantic search call takes ~200ms to embed your query. Subsequent calls within 5 minutes use cached data and are faster.