BrunoSan AI News MCP Server

Real-time AI news intelligence for AI agents · MCP-native · EU-hosted

What this is

A Model Context Protocol server that exposes the BrunoSan AI News intelligence layer to AI agents and MCP-compatible clients (Claude Desktop, Cursor, n8n, custom agents).

88,000+   AI and tech articles indexed
10,500+   curated global sources
77,000+   tracked entities (companies · people · products · regions)
        9 deterministic, auditable tools

Snapshot: 2026-05-03 · Articles archive reaches back to January 2020 · Daily growing.

Related MCP server: overtone-news-mcp

What this is not

This repository contains the read-layer that exposes the BrunoSan news database as MCP tools. Article ingestion, source curation, deterministic scoring, and clustering pipelines run upstream and remain operated by DoWell UG.

If you want the data, the recommended path is the hosted service. Self-hosting this server is supported but requires you to provide your own news database.

Use the hosted service

Free trial (50 calls, no card required) and paid tiers: https://brunosan.de/mcp.html

Endpoint: https://mcp.brunosan.de/mcp

After subscribing, you receive an API key. Pass it as the api_key parameter on every tool call. Works with any MCP-compatible client.

Claude Desktop

Add to your claude_desktop_config.json (see examples/claude_desktop_config.json):

{
  "mcpServers": {
    "brunosan-ai-news": {
      "url": "https://mcp.brunosan.de/mcp",
      "transport": "http"
    }
  }
}

Restart Claude Desktop. Ask: "What are today's trending AI topics?"

Smithery

Listed and verified: https://smithery.ai/server/@brunosan/global-ai-news

Self-host this server

For developers who want to point the server at their own news database, or who want to extend the read-layer.

Requirements

• Docker and Docker Compose

• A SQLite database with the BrunoSan news schema

• Python 3.12 if you prefer running without Docker

Quick start with Docker

git clone https://github.com/DowellUG/brunosan-ai-news-mcp.git
cd brunosan-ai-news-mcp
cp .env.example .env

# Place your news SQLite DB at:
mkdir -p data
cp /path/to/your/brunosan_news.db data/

docker compose up -d

The MCP endpoint will be available at http://localhost:8020/mcp.

To connect from Claude Desktop locally, point the config URL at your own host. To disable authentication entirely (development mode), set BRUNOSAN_AUTH_DB="" in the environment.

Run without Docker

pip install -r requirements.txt
export BRUNOSAN_NEWS_DB=/absolute/path/to/brunosan_news.db
export BRUNOSAN_AUTH_DB=""   # disable auth for local dev
python brunosan_news_mcp.py

The nine tools

All tools are read-only, idempotent, and require an api_key parameter.

Information tools

• news_search — Full-text search across all articles. Returns matched articles with title, teaser, source, URL, and publication date.

• news_get_entity — Profile of a company, person, product, or region. Returns recent articles and the entities most frequently co-mentioned alongside it.

• news_trending — Top topic clusters for a given day, ranked by burst score. Each cluster has a representative title, top entities, and counts.

• news_daily_digest — Full daily overview with statistics and top clusters. Designed for morning briefings and newsletter preparation.

• news_get_sources — Source statistics and database overview. Useful for transparency checks.

• news_search_entities — Discovery across known entities, with optional type filter (company, person, product, region).

Signal tools

• news_signal_strength — Seven deterministic 0–100 scores per article: AI relevance, business relevance, Mittelstand relevance, regulation, risk, cost, operational. All computed without LLMs and reproducible.

• news_decision_maker_brief — Cluster ranking for a chosen role (CDO, CTO, CEO, Compliance) using a transparent composite formula:

  composite_score = burst_score
                  × (tier1_source_count + 1)
                  × (hard_numbers_count + 1)
                  × ai_relevance_score

  The formula is identical across roles; the role only filters which clusters are eligible. The output includes every input component so the score can be verified.

• news_risk_radar — Risk and compliance signals ranked by an open formula:

  combined_risk_score = (risk_signal_count    × 2)
                      + (compliance_signal_count × 3)
                      + (official_source_count   × 1)

  Returns clusters sorted by combined score with all three component counts visible for audit.

Architecture

┌───────────────────┐       ┌──────────────────────┐       ┌─────────────────┐
│  MCP Client       │───────▶  This Server         │───────▶  News Database  │
│  (Claude/Cursor)  │  MCP  │  (read-only)         │ SQL   │  (read-only)    │
└───────────────────┘       └──────────────────────┘       └─────────────────┘
                                       │
                                       ▼
                            ┌──────────────────────┐
                            │  Auth (reference     │
                            │  or hosted version)  │
                            └──────────────────────┘

The server reads from news_articles, daily_clusters, daily_cluster_members, daily_cluster_evidence, daily_cluster_entities, article_entities, and article_fact_profile. It writes only to the auth-key database (call counters, when applicable).

Configuration

Production deployment notes

The hosted service uses Traefik for TLS termination and the multi-vertical auth library (per-vertical key scoping, tier limits, billing integration) which is operated separately and not in this repository.

For your own production deployment, terminate TLS with a reverse proxy (Caddy, Traefik, nginx) in front of this container, which speaks plain HTTP on port 8020.

License

Apache License 2.0 — see also NOTICE for attribution.

Operated by

DoWell UG · Hamburg, Germany

For partnerships, custom integrations, or workshop bookings: oliver@brunosan.de

Built daily, in the EU. 🐾