GreenCalculus
# GreenCalculus MCP server
[](https://github.com/greencalculus/greencalculus-mcp/actions/workflows/ci.yml)
[](https://www.npmjs.com/package/greencalculus-mcp)
[](https://registry.modelcontextprotocol.io)
[](#tools)
[](https://greencalculus.com/developers/)
[](./LICENSE)
Sourced greenhouse-gas emission factors and audit-traced carbon calculations, as an MCP server. Every value comes back with its exact source cell and a pinned data version — so an agent hands back a number a person can cite and a machine can reproduce, instead of a guess.
## Do you need this package?
Probably not. **The server is remote**, and if your client speaks remote MCP you should point it straight at the URL — nothing to install, nothing to update:
```json
{
"mcpServers": {
"greencalculus": {
"url": "https://mcp.greencalculus.com",
"headers": { "Authorization": "Bearer YOUR_KEY" }
}
}
}
```
This package exists for the clients that can only spawn a local stdio process, and for `docker run` installs. It is a thin bridge: it forwards each JSON-RPC message to the remote server and returns the reply verbatim. No method is special-cased, so new tools appear here without a release.
## Use it over stdio
```json
{
"mcpServers": {
"greencalculus": {
"command": "npx",
"args": ["-y", "greencalculus-mcp"],
"env": { "GREENCALCULUS_API_KEY": "YOUR_KEY" }
}
}
}
```
Or with Docker — `-i` is required and `-t` must be omitted, because the container's stdin/stdout *are* the transport and a TTY corrupts the stream:
```json
{
"mcpServers": {
"greencalculus": {
"command": "docker",
"args": ["run", "-i", "--rm", "-e", "GREENCALCULUS_API_KEY", "greencalculus/mcp"],
"env": { "GREENCALCULUS_API_KEY": "YOUR_KEY" }
}
}
}
```
Get a free key at **https://greencalculus.com/developers** — no card. Discovery (`initialize`, `tools/list`) works without one, and so do `search_factors` and `explain_absence`; every other tool needs a key.
## Tools
| Tool | What it does |
|---|---|
| `lookup_factor` | Fetch one emission factor by key, with its source and version |
| `lookup_factors` | Fetch many factors by key in one call — a portfolio is one request, not one per factor |
| `search_factors` | Search the corpus by free text |
| `resolve_factor` | Map a messy real-world description to the best-matching factor |
| `explain_absence` | Say *why* a factor does not exist, rather than returning nothing |
| `calculate_activity` | Activity → emissions, with unit conversion and GHG Protocol scope |
| `calculate_electricity` | Location-based and market-based electricity |
| `calculate_embodied` | Embodied carbon (EN 15978), explicit about missing lifecycle stages |
| `calculate_pcaf` | PCAF financed emissions, with the audit trail |
| `calculate_freight` | Freight by mode, distance and load |
| `calculate_spend` | Spend-based EEIO |
| `calculate_business_travel` | Business travel across modes |
## Configuration
| Variable | Default | Meaning |
|---|---|---|
| `GREENCALCULUS_API_KEY` | — | Your API key. `GC_API_KEY` is accepted as an alias; the explicit name wins. |
| `GREENCALCULUS_MCP_URL` | `https://mcp.greencalculus.com` | Override the endpoint. |
| `GREENCALCULUS_MCP_TIMEOUT_MS` | `120000` | Per-request timeout. |
Diagnostics go to stderr. Nothing but JSON-RPC is ever written to stdout — a stray byte there corrupts the session.
## Develop
```bash
npm test # unit tests, no network
node bin/greencalculus-mcp.js # reads JSON-RPC on stdin
docker build -t greencalculus/mcp .
```
## Releasing
Bump `version` in `package.json`, merge to `main`. That's the whole procedure.
[`release.yml`](.github/workflows/release.yml) asks npm and the MCP registry
whether they already have that version and publishes only where they don't, so
a merge that bumps ships it and a merge that doesn't is a no-op. It also runs
weekly, so a publish that failed is retried without a new commit.
`server.json` is the registry manifest, and the workflow rewrites its version
from `package.json` before publishing — one source of truth, three places that
have to agree.
npm authenticates by [trusted publishing](https://docs.npmjs.com/trusted-publishers),
so there is no npm token here. The registry needs one secret, and the reason is
worth knowing: we publish as `com.greencalculus/api`, a DNS namespace, and GitHub
OIDC only ever grants `io.github.<org>/*`. So the registry step signs with the key
matching the `v=MCPv1` TXT record on greencalculus.com, held as `MCP_PRIVATE_KEY`.
[`scripts/rotate-registry-key.sh`](./scripts/rotate-registry-key.sh) generates a
fresh pair and installs it without ever printing the private half; it prints the
TXT record to publish. Without the secret the registry step skips and says so —
npm still publishes.
## Also available
- **REST API** and docs — https://greencalculus.com/developers
- **Client SDKs** (Python, JS/TS) — https://github.com/greencalculus/greencalculus-sdk
- **Official MCP registry** — `com.greencalculus/api`
- **Smithery** — https://smithery.ai/servers/greencalculus/api
## Licence
MIT — see [LICENSE](./LICENSE). The licence covers this bridge. Emission-factor data returned by the API carries the licence of its underlying source, which is named in every response.
TDQS
Scored across 12 tools
The seven calculate_* tools map cleanly to distinct emissions categories, and the four factor-retrieval tools (search_factors, resolve_factor, lookup_factor, lookup_factors) are explicitly cross-referenced to guide selection. Still, search_factors and resolve_factor both 'find' factors, and lookup_factor vs lookup_factors differ only by cardinality, leaving some borderline overlap an agent must read carefully to avoid misselecting.
Every tool follows a predictable verb_noun snake_case pattern: search_factors, resolve_factor, lookup_factor(s), calculate_activity/embodied/pcaf/electricity/freight/spend/business_travel, explain_absence. No mixed conventions, so the naming is highly predictable.
Twelve tools is well-scoped for an emissions-factor and GHG-calculation server, with each calculate_* earning its place and the retrieval tools covering search, resolve, single, and batch lookup. Nothing feels padded or thin.
The surface covers the full workflow: find/verify/resolve factors, compute across multiple Scope 1-3 and lifecycle methods, and explain gaps via absence classification. Minor omissions exist (e.g. some Scope 3 categories like waste/water and any write/update capability), but core GHG accounting workflows reach no dead ends.