Nordic Financial MCP

<!-- mcp-name: io.github.AIDataNordic/nordic-financial-mcp -->

A production-grade semantic search server for Nordic financial markets — built for autonomous AI agents. 1,000,000+ vectors across exchange filings, company reports, commodity prices, freight rates, energy data and press releases.

Search: Natural language queries over annual reports, quarterly reports, exchange announcements and macroeconomic summaries — filtered by company, ticker, country, sector or year. Two-stage hybrid retrieval (dense + sparse BM25, fused via RRF) with cross-encoder reranking for high-precision results.

Live endpoint: https://mcp.aidatanorge.no/mcp Transport: streamable-http Registry: Smithery · MCP Registry · Glama · mcp.so

Connect

Add to your MCP client config:

{
  "mcpServers": {
    "nordic-financial": {
      "type": "streamable-http",
      "url": "https://mcp.aidatanorge.no/mcp"
    }
  }
}

Or with Claude Code:

claude mcp add --transport http nordic-financial https://mcp.aidatanorge.no/mcp

Quick Test

Try the live demo in your browser: 👉 https://mcp.aidatanorge.no/demo

No installation, no configuration. Just search for "Equinor dividend", "Swedish policy rate", or "salmon price Q3".

For MCP Client Developers

This server follows the StreamableHTTP MCP transport. A complete handshake is required before calling tools.

Full Handshake Example (Copy-Paste Ready)

# 1. Create session and capture session ID
SESSION_ID=$(curl -X GET https://mcp.aidatanorge.no/mcp \
  -H "Accept: application/json, text/event-stream" \
  -s -i | grep -i "mcp-session-id" | awk '{print $2}' | tr -d '\r')

echo "Session ID: $SESSION_ID"

# 2. Initialize session
curl -X POST https://mcp.aidatanorge.no/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "mcp-session-id: $SESSION_ID" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "initialize",
    "params": {
      "protocolVersion": "2024-11-05",
      "capabilities": {},
      "clientInfo": {"name": "example-client", "version": "1.0"}
    }
  }'

# 3. Send initialized notification
curl -X POST https://mcp.aidatanorge.no/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "mcp-session-id: $SESSION_ID" \
  -d '{"jsonrpc": "2.0", "method": "notifications/initialized"}'

# 4. List available tools
curl -X POST https://mcp.aidatanorge.no/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "mcp-session-id: $SESSION_ID" \
  -d '{"jsonrpc": "2.0", "id": 2, "method": "tools/list", "params": {}}'

# 5. Perform a search
curl -X POST https://mcp.aidatanorge.no/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "mcp-session-id: $SESSION_ID" \
  -d '{
    "jsonrpc": "2.0",
    "id": 3,
    "method": "tools/call",
    "params": {
      "name": "search_filings",
      "arguments": {"query": "Equinor dividend", "limit": 3}
    }
  }'

Common Issues & Solutions

Python Example with MCP SDK

from mcp import ClientSession
from mcp.client.streamable_http import streamablehttp_client

async with streamablehttp_client("https://mcp.aidatanorge.no/mcp") as transport:
    async with ClientSession(*transport) as session:
        await session.initialize()
        tools = await session.list_tools()
        result = await session.call_tool(
            "search_filings",
            {"query": "Norwegian housing market Q3 2024", "country": "NO"}
        )
        print(result.content[0].text)

Why This Matters

This handshake is automatic in MCP-compliant clients like Claude Desktop, LangChain, and the MCP Python SDK. If you're building a custom client, following the sequence above ensures compatibility.

The /demo endpoint shows how a browser can perform the same handshake using JavaScript fetch() — view source for a working implementation.

What This Is

AIDataNorge is a full-stack data pipeline and semantic search system that ingests, processes, and indexes financial data from Nordic markets into a vector database optimized for AI agent queries. It exposes data through a Model Context Protocol (MCP) server, making it natively compatible with Claude, LangChain, and other LLM-based agents.

The system is designed with autonomous machine-to-machine consumption in mind, including support for emerging agent payment protocols. The database is updated nightly.

MCP Tools

search_filings

Semantic search over Nordic company filings, press releases and macroeconomic summaries.

search_filings(
    query="Nordea net interest margin outlook 2025",
    report_type="quarterly_report",  # annual_report | quarterly_report | press_release | macro_summary
    country="SE",                    # NO | SE | DK | FI
    ticker="NDA",                    # optional — filter by company ticker
    fiscal_year=2025,                # optional — filter by year
    sector="energy",                 # optional — seafood | energy | shipping
    limit=10                         # default 5, max 20
)
# Returns semantically ranked text chunks with rerank_score, hybrid_score, vector_score,
# company, ticker, country, fiscal_year, report_type, filing_date and full text.

Search pipeline: Dense embedding (intfloat/e5-large-v2, 1024d) + sparse BM25, fused via Reciprocal Rank Fusion (RRF), reranked by mmarco-mMiniLMv2-L12-H384-v1. Natural language queries in any language are supported.

get_company_info

Look up a company in the official business registry.

get_company_info(
    identifier="923609016",  # org/CVR/business ID
    country="NO"             # NO (Brønnøysund) | DK (CVR) | FI (PRH)
)
# Returns company name, status and registered address.

parse_pdf_to_text

Download a PDF from a URL and extract all text, page by page.

parse_pdf_to_text(
    pdf_url="https://example.com/annual_report_2024.pdf"
)
# Returns extracted text with page separators.
# Useful for reading report attachments not indexed in the main database.

get_current_power_price

Real-time day-ahead electricity spot prices for all Nordic bidding zones.

get_current_power_price(
    zone="NO1",              # NO1–NO5, SE1–SE4, DK1, DK2, FI
    include_tomorrow=False   # fetch tomorrow's prices if available (published ~13:00 CET)
)
# Returns EUR/kWh — current hour price + full hourly breakdown + daily min/max/avg.
# Norwegian zones sourced from hvakosterstrommen.no, others directly from ENTSO-E.
# Handles both PT60M (hourly) and PT15M (15-min) resolutions.

company_research

Run multiple targeted searches in a single call and get raw results grouped by section. The caller defines all sections and queries and is responsible for synthesizing the output.

company_research(
    company="Equinor",
    sections=[
        {"name": "financials", "query": "Equinor revenue EBITDA operating profit 2024", "ticker": "EQNR"},
        {"name": "risk",       "query": "Equinor climate regulatory risk stranded assets", "ticker": "EQNR"},
        {"name": "macro",      "query": "Brent crude oil price energy sector Norway 2024", "limit": 3},
        {"name": "news",       "query": "Equinor press release dividend acquisition 2024", "ticker": "EQNR"}
    ]
)
# Returns: {company, generated_at, sections} — one entry per section with ranked text chunks.
# All sections are searched in parallel. Up to 8 sections, max 10 results each.
# Use ticker on company-specific sections to avoid false positives from documents
# that merely mention the company as a customer or competitor.

For a fully orchestrated due diligence report where AI plans the sections and synthesizes the narrative, use Alfred MCP instead.

ping

ping(name="world")
# Returns: "Hello world! Nordic MCP server is running."

Data Coverage

Total: 1,000,000+ vectors · Updated nightly

Architecture

Data Sources                 Pipeline                  Serving
─────────────────            ─────────────────         ─────────────────
XBRL ESEF               →    Python ingest scripts  →  Qdrant
MFN Nordics             →    + Playwright scraping  →  Vector Database
Oslo Børs Newsweb       →    + PDF extraction        →  (1,000,000+ vectors)
Nasdaq Copenhagen       →    + Chunking              →        ↓
Cision / GlobeNewswire  →
SSB / Norges Bank       →    + Chunking              →        ↓
SSB / Norges Bank       →    + Dense embeddings      →  MCP Server
SCB / DST / stat.fi     →
                        →      (e5-large-v2, 1024d)  →  (FastMCP 3.2)
                        →    + Sparse BM25            →        ↓
                        →    + RRF fusion             →  AI Agents / LLMs

Technical Stack

Data ingestion

• Python with Playwright for JavaScript-rendered IR pages and MFN feed
• PyMuPDF (fitz) for PDF text extraction
• Paragraph-aware chunking (512-token chunks, 100-token overlap)
• Dense embeddings: intfloat/e5-large-v2 (1024d)
• Sparse embeddings: Qdrant/bm25 via fastembed

Storage & search

• Qdrant vector database (self-hosted)
• Hybrid dense+sparse retrieval with Reciprocal Rank Fusion (RRF)
• Cross-encoder reranking (mmarco-mMiniLMv2-L12-H384-v1)

Serving

• FastMCP 3.2 over HTTP (/mcp endpoint)
• Cloudflare Tunnel — rate limited to 60 req/min per IP
• Compatible with Claude, LangChain, and any MCP-capable agent

Infrastructure

• Ubuntu Server 24 LTS, self-hosted
• 16 GB RAM
• Automated cron jobs for continuous ingestion
• Bitcoin full node (LND) for Lightning Network payments
• DigiByte full node with DigiRail and DigiDollar Oracle node

Agent Payment Infrastructure

The system is built with autonomous agent monetization in mind, supporting three complementary payment protocols:

x402 Micropayments A pay-per-call variant of the server (mcp_server_x402.py) is implemented using the x402 protocol — the HTTP 402 payment standard for autonomous agents. Agents receive a payment requirement response, pay in USDC on Base, and retry automatically. Currently paused — x402 functionality will be integrated directly into the main server (mcp_server.py) in a future release.

Lightning Network (L402) Running a full Bitcoin node with LND enables L402 — the HTTP payment protocol for autonomous agents. Agents can discover the API, receive a Lightning invoice, pay in millisatoshis, and get access — all without human intervention. Infrastructure in place, monetization layer in development.

DigiRail / DigiDollar Also running a DigiByte full node with DigiRail (an agent payment protocol similar to L402) and a DigiDollar Oracle node. DigiDollar is the world's first UTXO-native decentralized stablecoin, implemented directly in DigiByte Core v9.26. The oracle node contributes to the decentralized price feed that maintains DigiDollar's USD peg — 15 of 30 randomly selected oracle nodes must reach consensus every ~25 minutes using Schnorr signatures.

This multi-protocol payment infrastructure (x402/Base + Bitcoin/Lightning + DigiByte/DigiRail) positions AIDataNorge to serve agents operating across different payment ecosystems.

Ingest Pipeline Design

Each data source has a dedicated ingest script with:

• Idempotent processing via MD5-based point IDs (upsert-safe)
• processed.txt log to avoid redundant re-fetching
• nohup + cron scheduling for unattended overnight runs
• Structured payload per chunk: source, country, ticker, company_name, report_type, published_date, chunk_index, total_chunks

Chunking strategy: paragraphs are accumulated until reaching the 512-token model window. Chunks never split mid-sentence. 100-token overlap ensures context continuity across chunk boundaries.

Cron Schedule

Monitoring & Activity

Check server health

# Qdrant responding?
curl http://localhost:6333

# Vector count
curl http://localhost:6333/collections/nordic_company_data | python3 -m json.tool

Check MCP server process

ps aux | grep mcp_server.py

Check MCP query activity

# Tail live log
tail -f ~/logs/mcp_server.log

# Run full query analysis
cd ~/norsk-mcp-server && venv/bin/python3 analyze_queries.py

Check Cloudflare tunnel

journalctl -u cloudflared --since "1 hour ago" | tail -50

Skills Demonstrated

• RAG system design — end-to-end pipeline from raw data to semantic search
• Hybrid retrieval — dense+sparse embeddings with RRF fusion and cross-encoder reranking
• Web scraping at scale — Playwright, RSS feeds, REST APIs, PDF extraction
• Vector database operations — Qdrant, embedding models, reranking
• MCP server development — FastMCP, tool design for LLM agents
• Agent payment protocols — x402, L402, DigiRail
• Linux server administration — process management, cron, systemd
• Blockchain infrastructure — Bitcoin full node + LND, DigiByte full node + oracle
• Python engineering — async pipelines, error handling, idempotent design
• Financial data domain knowledge — Nordic exchanges, regulatory filings, macro data

Status (May 2026)

• nordic_company_data: 1,000,000+ vectors — XBRL, MFN, Newsweb, Cision, GlobeNewswire, ENTSO-E, commodity/freight, macro
• MCP server: live at https://mcp.aidatanorge.no/mcp
• Published: Smithery · MCP Registry · Glama · mcp.so
• x402 pay-per-call: implemented, currently paused — will be integrated into main server
• L402 / DigiRail: infrastructure in place, monetization layer in development
• Live demo: https://mcp.aidatanorge.no/demo