Summary
The four steps SDMX makes you take, made askable.
Getting one IMF series normally means finding the database, then its dimensions, then the codes, then writing the query. All four are tools here, so the discovery happens in the conversation rather than in the API docs.
What it is
A query path into the IMF's SDMX API at data.imf.org, built on imfp. Roughly 155 databases are searchable, each one's dimensions and valid codes are discoverable, and the series comes back filtered by year range and dimension.
What you get
- All ~155 IMF databases listed with their ids, or filtered by keyword such as inflation or trade
- The query dimensions a given database accepts — country, frequency, index type and the rest
- Valid codes for a dimension, searchable, so a country name becomes the code the API expects
- Time series fetched with a start and end year and dimension filters applied
- Rate limiting and retries handled underneath, against an API that allows roughly 10 requests / 5 seconds
Requirements
Python with uv or pip; uvx runs it straight from the repository. No account and no key. A recent release moved off dataservices.imf.org, which the IMF has decommissioned — earlier versions time out on every request.
Setup effort
One command — uvx --from git+https://github.com/c-cf/imf-data-mcp imf-data-mcp