energy-emissions-mcp
# energy-emissions-mcp
An MCP server that exposes US energy and emissions data as tools an AI agent can call.
It currently has one tool, `get_facility_emissions`, which answers questions like
"who are the biggest emitters in Harris County, TX?" from the EPA Greenhouse Gas
Reporting Program (GHGRP), queried live through the
[Envirofacts API](https://www.epa.gov/enviro/envirofacts-data-service-api). No API key needed.
## Run it
Requires [uv](https://docs.astral.sh/uv/).
```sh
uv sync
uv run mcp dev server.py # opens MCP Inspector to call the tool by hand
uv run pytest # offline tests with recorded EPA rows
```
Add it to Claude Code:
```sh
claude mcp add energy-emissions -- uv run --directory "$PWD" mcp run server.py
```
## The tool
`get_facility_emissions(state, county=None, year=2023, limit=25)` returns each facility's
name, city, county, industry and total CO2e (metric tons), largest first, plus the place's
total, the source tables and the exact query URL so answers can cite it.
Totals are direct emissions, matching GHGRP's own facility totals: biogenic CO2 and
fuel suppliers' downstream emissions are left out. See [docs/ghgrp-notes.md](docs/ghgrp-notes.md).
Built with the [MCP Python SDK](https://py.sdk.modelcontextprotocol.io/) v2 and httpx.
TDQS
Scored across 1 tool
There is only one tool, so there is no possibility of misselection between tools. Its description also clearly delimits scope (GHGRP facility-level emitters, not a place's total emissions footprint), preventing misuse.
The single name `get_facility_emissions` follows a clean verb_noun snake_case convention that would scale predictably. With only one tool there is no cross-tool consistency to verify, so it falls short of a perfect score.
One tool is very thin for a whole emissions data server; there is no way to discover valid states/counties/years or drill into a specific facility. The single endpoint is well-scoped for its narrow query, but the surface is minimal for the apparent domain.
The core query (facility emissions by state/county/year) is well covered, including sorting, limits, totals, and citation metadata. However, notable gaps remain: no discovery of valid states/counties/years, no facility-level detail lookup, and no comparison or trend operations.