SwemoMCO is an unofficial Monetary‑Policy Data MCP Server wraps Sveriges Riksbank’s open API in a Model‑Context‑Protocol (MCP) micro‑service. It turns the raw REST end‑points into typed Python tools that can be invoked by LLMs or by humans through any MCP Client.
This edition of the README assumes you are using Astral’s uv for dependency management and execution.
- Why this exists
- The underlying data
- How the MCP server works
- Installation
- Quick‑start
- Examples (Analysts & Journalists)
- Docker
- Development
- License
Working with monetary‑policy data can be complicated: the raw API requires hand‑crafted queries and knowledge of the different series. This project:
- hides the HTTP plumbing behind a clean, async Python interface,
- exposes every series as an MCP tool discoverable by LLMs – allowing Claude Desktop and other MCP client tools to fetch Swedish macro data on demand.
As of 2020, the Riksbank’s service now includes both forecast values and realised (observed) data once official figures are published. This makes the dataset suitable for historical analysis (e.g. “What actually happened to inflation in 2022?”) and for forecast queries (e.g. “What does the Riksbank project for GDP next year?”).
The Riksbank publishes a fresh set of forecasts four to five times a year.
Each publication is labelled YYYY:I (e.g. 2025:2 for the second release in 2025).
Every time‑series name follows the pattern
COUNTRY‑FREQUENCY‑AREA‑DECOMPOSITION‑UNIT‑ADJUSTED
Example: SEQGDPNAYCA → Sweden (SE), Quarterly (Q), GDP (GDP),
National‑Accounts decomposition (NA), y/y change (Y), Calendar
adjusted (CA). Discover the catalogue with
GET /forecasts/series_idsEach policy round gives rise to a new “vintage” of forecasts for key macroeconomic variables. Meanwhile, as data on actual outcomes get published, the Riksbank updates its realised observations. This means you can:
- Pin a specific policy round (e.g.
"2024:1") to see only the forecasts from that round along with any observations available up through that round. - Use
"latest"to retrieve all historical observations (the final realised values known today) plus the newest forecast vintage. This is ideal for historical analysis—especially if you want to see the final, revised or actual values rather than the older forecasts.
Forecast metadata example:
┌─────────────────────────────────────────────────────────┐
│ FastMCP Server (src/swemo_mcp/server.py) │
│ │
│ • registers ≈30 *tools* (one per economic series) │
│ • exposes them on stdio / SSE / HTTP │
└───────────────▲──────────────────────────────▲──────────┘
│ │
async httpx LLM / user
│ │
┌───────────────┴────────────┐ ┌────────┴──────────┐
│ Riksbank REST API │ │ mcp‑cli / ChatGPT │
└────────────────────────────┘ └───────────────────┘
services/monetary_policy_api.py– thin async wrapper with automatic exponential back‑off (max‑retry on HTTP 429).tools/monetary_policy_tools.py– one declarative function per series; docstrings double as LLM prompts.- Pydantic v2 models in
models.pyensure every response has the expected schema.
| Tool | Series ID | Description |
|---|---|---|
get_gdp_data |
SEQGDPNAYCA |
GDP y/y, calendar‑adjusted |
get_unemployment_data |
SEQLABUEASA |
LFS unemployment rate |
get_cpi_data |
SEMCPINAYNA |
Headline CPI y/y |
| … | … | ≈ 30 series in total – run list_series_ids() for the full list. |
Each tool signature is:
async def get_<series>_data(policy_round: str | None = None) -> MonetaryPolicyDataResponsePass policy_round="2024:3" to pin the vintage; omit for the complete
history. For final historical data, pass policy_round="latest" so that
the tool merges all realised (observed) data points.
Prerequisites:
- Python ≥ 3.12 (uses
typing.TypeAlias/PEP 604 unions)- Astral
uv≥ 0.2.0
Clone and set up the project with one command:
uv syncuv sync installs all production and development dependencies declared in
pyproject.toml, creates a virtual environment if needed, and locks the
exact versions so every contributor or CI pipeline uses the same stack.
Edit your claude_desktop_config.json to add Kolada MCP Server:
"SwemoMCP": {
"args": [
"run",
"-i",
"--rm",
"--name",
"swemo-mcp-managed",
"swemo-mcp:local"
],
"command": "docker",
"env": {}
}"SwemoPyPI": {
"args": ["swemo-mcp"],
"command": "/Users/hugi/.cargo/bin/uvx"
}Replace [path to kolada-mcp] with your local directory:
"SwemoLocal": {
"args": [
"--directory",
"[path to kolada-mcp]/src/kolada_mcp",
"run",
"kolada-mcp"
],
"command": "uv"
}Restart Claude Desktop after updating.
import asyncio
from swemo_mcp.tools import get_policy_rate_data
async def main():
from swemo_mcp.query import ForecastRequest
req = ForecastRequest(policy_round="2023:4", include_realised=True)
data = await get_policy_rate_data(req)
print(data.vintages[0].observations[:5]) # first 5 observations
asyncio.run(main())Because everything is typed and async, you can integrate the tools directly into notebooks, dashboards, or other services.
The project ships with a multi‑stage Dockerfile that uses uv in the final
layer, so container builds benefit from deterministic dependency resolution.
docker build -t swemo-mcp:latest .
docker run -i --rm swemo-mcp:latest | mcp chatIf you prefer Docker Compose for development, a sample compose.yaml
illustrates how to mount the source directory and hot‑reload changes.
-
Set up the environment:
uv sync --dev
-
Run the server in dev‑mode with live‑reload (requires
mcp dev):uv run mcp dev src/swemo_mcp/server.py
-
Open the MCP Inspector to test and debug:
-
Run the test‑suite (pytest + asyncio):
uv run pytest -q
-
Format & lint automatically with Ruff:
uv run ruff check . --fix
Licensed under the Apache 2.0 license. See LICENSE for the
full text.
- This is an unofficial MCP server for the Riksbank’s data. The underlying API is subject to change, and this project may not always reflect the latest updates.
- Sveriges Riksbank has had no involvement in the development of this project. The data is provided "as is" without any warranty of any kind.
- The Riksbank’s data is subject to its own terms of use. Please refer to the Riksbank’s API portal for more information.
{ "revision_dtm": "2024‑07‑02T08:55:00Z", "forecast_cutoff_date": "2024‑06‑18", "policy_round": "2024:3", "policy_round_end_dtm": "2024‑07‑02T09:30:00Z" }