Skip to main content
Glama
Nordic Financial MCP

Nordic Financial MCP

by AIDataNordic

Nordic Financial MCP

nordic_financial_mcp MCP server

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

Related MCP server: SEC EDGAR MCP Server

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

Error

Cause

Solution

406 Not Acceptable

Missing text/event-stream in Accept header

Send: Accept: application/json, text/event-stream

400 Bad Request: Missing session ID

No session established

First GET /mcp, use returned mcp-session-id header

-32602 Invalid request parameters

Missing initialize before tools/list

Complete steps 1-3 in order

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.

due_diligence_report

Run multiple targeted searches in a single call and get results grouped by section — designed for agent-driven due diligence workflows.

due_diligence_report(
    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.

The agent defines all sections and queries — the tool does not decide what is relevant. Reason about which topics matter for the specific company before calling: financial metrics, risk factors, sector-specific macro drivers, recent press releases, peer context.

ping

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

Data Coverage

Source

Geography

Content

Volume

XBRL ESEF (filings.xbrl.org)

NO/SE/DK/FI/IS

Annual reports, regulated markets, 2020–present

~89k vectors

MFN Nordics

SE/NO/DK/FI

Annual & quarterly reports, First North companies

~116k vectors

Oslo Børs Newsweb

NO

Exchange announcements, 2020–present

~52k vectors

Nasdaq Copenhagen

DK

Exchange announcements, 2020–present

~8k vectors

Nasdaq Helsinki

FI

Exchange announcements, 2020–present

~5k vectors

Nasdaq Stockholm

SE

Exchange announcements, 2020–present

in progress

Cision

SE/NO/DK/FI

Press releases

~20k vectors

GlobeNewswire

NO/SE/DK/FI

Press releases, updated hourly Mon–Fri

~500 vectors

ENTSO-E

NO/SE/DK/FI

Day-ahead electricity prices, all bidding zones

~24k vectors

Commodity & freight

Global

Oil, gas, metals, shipping rates (BDRY/FRO/ZIM proxies)

25 quarters

Macro Norway

Norway

GDP, CPI, rates, housing, salmon, power

24 quarters

Macro Nordics

SE/DK/FI

Rates, housing, credit, power

72 quarters

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

Time

Job

03:17 Sundays

XBRL annual reports

06:00 Mon–Fri

yfinance — stock prices and FX rates

06:15 daily

MFN Nordics — quarterly reports and press releases

06:30 Mon–Fri

ENTSO-E — energy data

07:00 daily

Oslo Børs Newsweb — exchange announcements

08:00–18:00 hourly Mon–Fri

GlobeNewswire — press releases (NO/SE/DK/FI)

09:00 daily

Query analysis report (email)


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

Available Tools

6 tools
due_diligence_reportA
Read-only
Inspect

Run multiple targeted searches and return results grouped by section for due diligence.

The agent defines all sections and queries — this tool does not decide what is relevant. Before calling, reason about which topics and data sources matter for this specific company: financial metrics, risk factors, sector-specific macro drivers (e.g. freight rates for shipping, power prices for aluminium smelters), recent press releases, peer context, etc. Formulate one query per section.

Each query is run independently as a full hybrid search (dense + sparse + rerank).

IMPORTANT — use 'ticker' on company-specific sections to avoid false positives. Without a ticker filter, documents that merely mention the company (e.g. as a customer or competitor) can rank above actual filings from that company. Omit 'ticker' only for sections where cross-company results are intentional, such as sector macro context or peer comparisons.

Inputs: company (required), sections (required). Returns: object

get_company_infoA
Read-only
Inspect

Look up a company in the official business registry for Norway, Denmark or Finland.

Use this to retrieve authoritative registration data (legal name, status, address) for a known organisation number. Do not use for Sweden (SE) — use search_filings with country='SE' instead, as Bolagsverket integration is not yet available. Do not use to discover tickers or ISIN codes — use search_filings for that.

Inputs: identifier (required), country. Returns: object

get_current_power_priceA
Read-only
Inspect

Fetch today's hourly day-ahead electricity spot prices for a Nordic bidding zone.

Use this for current and near-term (today/tomorrow) price queries. Do not use for historical price analysis — use search_filings with report_type='macro_summary' and a date reference in the query for that purpose. Tomorrow's prices are published by NordPool around 13:00 CET; requests before that time will return "not yet available" for the tomorrow field.

All zones return prices in EUR/kWh (NordPool day-ahead, native currency). Norwegian zones (NO1–NO5) use hvakosterstrommen.no; all other zones use ENTSO-E.

Inputs: zone, include_tomorrow. Returns: object

parse_pdf_to_textA
Read-only
Inspect

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

Use this to read the full text of a specific document — for example, an annual report PDF linked from a search_filings result. Best combined with search_filings: use search_filings to locate the document, then parse_pdf_to_text for the full text. Do not use for PDFs that are already well-represented in the database — search_filings is faster and returns pre-ranked, relevant excerpts. Not suitable for scanned (image-only) PDFs without embedded text; those pages will be returned as "(no extractable text)".

Inputs: pdf_url (required). Returns: result

pingA
Read-only
Inspect

Connectivity check that confirms the Nordic MCP server process is responding.

Use this at the start of a session to verify the server is reachable before making other calls. Do not use as a proxy for database health — the server can respond while the Qdrant vector database is temporarily unavailable. To confirm data availability, call search_filings directly.

Returns: A greeting string: "Hello {name}! Nordic MCP server is running."

Inputs: name. Returns: result

search_filingsA
Read-only
Inspect

Search the Nordic financial database for company filings, press releases and macroeconomic summaries.

Use this as the primary tool for any question about Nordic listed companies, markets or macro conditions. Do not use to retrieve a full document — results are chunked text excerpts; use parse_pdf_to_text for the full original document. Do not use for Swedish company registration data — use get_company_info instead.

The database contains ~1 million vectors across four Nordic markets (NO/SE/DK/FI).

COMPANY FILINGS Annual reports (XBRL/ESEF) and quarterly reports from ~1 500 listed companies across Oslo Børs, Nasdaq Stockholm, Nasdaq Helsinki, Nasdaq Copenhagen and First North markets. Covers 2020–present. Strong coverage for NO and SE; growing coverage for DK and FI.

EXCHANGE ANNOUNCEMENTS & PRESS RELEASES Regulatory filings, exchange announcements and press releases from listed companies in NO, SE, DK and FI. Covers 2020–present.

MACROECONOMIC SUMMARIES Quarterly macro summaries covering key indicators per country: Norway (NO): policy rate, FX rates, CPI, house prices, credit growth, electricity price, salmon price, GDP components Sweden (SE): policy rate, house price index, household credit Denmark (DK): policy rate, house price index, household loans, electricity price Finland (FI): house price index, household debt-to-income ratio, electricity price Use report_type='macro_summary' and country='NO'/'SE'/'DK'/'FI' to filter. Use fiscal_year and a quarter reference in your query, e.g. "Norwegian housing market Q1 2024".

Inputs: query (required), ticker, fiscal_year, report_type, +3 more. Returns: result

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.2/5.0

    Scored across 6 tools

    Disambiguation2/5

    get_company_info and search_filings have contradictory scoping: get_company_info tells you to use search_filings for Swedish company data, while search_filings tells you to use get_company_info instead. Additionally, due_diligence_report is essentially a multi-query wrapper around the same search functionality, creating a boundary question for agents.

    Naming Consistency4/5

    Most tools follow a clear verb_noun pattern (get_company_info, search_filings, parse_pdf_to_text, get_current_power_price). The exceptions are ping, which is a standard health-check convention, and due_diligence_report, which is noun-oriented but still recognizable. Overall readable and mostly predictable.

    Tool Count5/5

    Six tools is a well-scoped count for the described domain: connectivity, structured registry lookup, broad search, PDF processing, a market-specific data tool, and a multi-step report generator. Each tool earns its place without redundancy or bloat.

    Completeness3/5

    The surface covers search, company info, full-text PDF access, power prices, and report generation, but there are notable gaps: no structured Swedish company registry support (despite Nordic coverage), no direct way to list all filings for a given company beyond hit-based search, and no tool for retrieving historical power prices except via macro summaries. These gaps can likely be worked around but may cause agent uncertainty.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • F
      license
      Not graded
      quality
      Not graded
      maintenance
      Provides comprehensive Norwegian business intelligence through Brønnøysund and Statistics Norway APIs, enabling company search, financial analysis, ownership mapping, market research, and automated financial data extraction.
      -
    • F
      license
      A
      quality
      D
      maintenance
      Provides access to SEC filings and detailed XBRL financial data for all publicly traded U.S. companies. It enables users to search for company info, retrieve historical metrics like revenue and assets, and compare financial performance across different industries.
      6
      1
      -
    • A
      license
      A
      quality
      D
      maintenance
      Provides access to a suite of Nordic data tools covering Danish business records, addresses, weather, and energy prices, alongside Norwegian and Finnish company information. This unified server enables users to query public APIs for regional data across Denmark, Norway, and Finland without requiring individual API keys.
      33
      12 npm
      2
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Financial data and research MCP for AI agents: filings with full-text and fact search, statements as reported, earnings, insider and institutional ownership, corporate events, executives, analyst data, company discovery and research signals for US, China and Japan equities. Every figure traced to its filing. Browser sign-in.
      10
      MIT