Lead Green
GPU energy per million output tokens by model, with carbon estimates by country
- Pricing
- Open source
- Platforms
- CLI, macOS, Linux
- Runs
- Locally
- Version
- 0.3.1
- Listing updated
- 2 days ago
About Lead Green
Lead Green is an open-source MCP server for ai models and llm tooling. It runs locally on your machine and is started by your AI client.
Answers come from the Lead Green index, a public dataset of the watt-hours needed to serve a million output tokens, by model, GPU and configuration. The README says every figure traces to a named public source with a retrieval date, and that the numbers are GPU-only floors measured on a busy server. Every response says whether it used the live API, a local folder or the bundled snapshot, and when the data was last updated.
The estimate tool turns a token count into Wh and kWh and, given a country, kg of CO2e. Optional inputs set the running batch, the number of GPUs, a latency ceiling, and a PUE and system power ratio to estimate energy at the meter; those two are your own assumptions, not measurements. Other tools compare H100 with B200, show energy at each measured batch size, and report grid intensity and GPU hardware data.
It runs locally over stdio and only makes plain HTTPS GET requests to leadgreen.ai. Your questions and token counts never leave your machine, and pointing LEADGREEN_API_BASE at a local folder removes network use entirely. Python 3.10 or later is required.
Install
Run in a terminal
claude mcp add green-token-index -- uvx leadgreen-mcpAdd to claude_desktop_config.json
{
"mcpServers": {
"green-token-index": {
"command": "uvx",
"args": [
"leadgreen-mcp"
]
}
}
}Add to ~/.cursor/mcp.json
{
"mcpServers": {
"green-token-index": {
"command": "uvx",
"args": [
"leadgreen-mcp"
]
}
}
}Add to .vscode/mcp.json
{
"servers": {
"green-token-index": {
"type": "stdio",
"command": "uvx",
"args": [
"leadgreen-mcp"
]
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"green-token-index": {
"command": "uvx",
"args": [
"leadgreen-mcp"
]
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"green-token-index": {
"command": "uvx",
"args": [
"leadgreen-mcp"
]
}
}
}Tools this server provides (8)
Taken from the project documentation; the running server may differ.
- list_models
- Lists models on the index with slugs, parameter counts and measured GPUs.
- energy_per_million_tokens
- Returns index rows for one model, optionally one GPU.
- estimate
- Estimates energy and carbon for a token count.
- compare_gpus
- Compares H100 and B200 for a model.
- concurrency_curve
- Shows energy per token at each measured running batch.
- grid_intensity
- Returns annual average gCO2e per kWh for a country.
- gpu_hardware
- Returns TDP, memory, MIG profiles and system ceilings.
- sources
- Lists every source with version, retrieval date and licence.
What it does
- Energy per token lookup
- Returns the index rows for a model, optionally for one GPU, within a latency ceiling.
- Workload estimates
- Converts a token count into Wh, kWh and kg CO2e for a chosen country.
- GPU comparison
- Compares H100 and B200 for a model and reports the ratio.
- Concurrency curve
- Shows energy per token at every measured running batch.
- Grid and hardware data
- Gives annual average gCO2e per kWh by country and GPU TDP, memory and MIG details.
- Offline fallback
- Uses a bundled snapshot when the site is unreachable and labels the data source.
Good for
- Estimate the energy and carbon of serving 50 million tokens of a given model
- Compare B200 and H100 for a specific model's energy use
- Check the grid carbon intensity for a deployment country
- Cite the source and caveat behind an AI energy figure
Pricing
Open source
Apache 2.0 package; data under CC BY 4.0
Prices change. Check leadgreen.ai before you buy.
Lead Green FAQ
Does it send my queries anywhere?
No. Your questions, token counts and countries stay on your machine and the arithmetic runs locally. It only fetches published JSON files from leadgreen.ai.
Can it work offline?
Yes. It falls back to the snapshot bundled in the package, or you can point LEADGREEN_API_BASE at a local folder of API files.
What is the licence?
The package is Apache 2.0 and the data is published under CC BY 4.0.
How do I install it?
Run pip install leadgreen-mcp or uvx leadgreen-mcp, then add the leadgreen-mcp command to Claude Desktop, Claude Code or Cursor.
Is Lead Green free?
Yes. Lead Green is open source, so you can use and self-host it for free.
How do I add Lead Green to Claude Code?
Run this command in your terminal: claude mcp add green-token-index -- uvx leadgreen-mcp
What are the best alternatives to Lead Green?
The closest alternatives to Lead Green are Agent-Native Analytics, Claude Code Ultimate Guide, Google Ads + Meta Ads + SEO/GEO MCP, Tapetide - Indian Stock Market MCP and World Monitor.
Alternatives to Lead Green

Agent-Native Analytics
MCP serverOpen-source analytics app where an AI agent builds charts and dashboards from your data
- Remote
- Open source
- 7.7K GitHub stars
- Analytics and data

Claude Code Ultimate Guide
MCP serverMCP server for searching the Claude Code Ultimate Guide from any MCP client
Runs locally- Open source
- 6.5K GitHub stars
- AI models and LLM tooling

Google Ads + Meta Ads + SEO/GEO MCP
MCP serverHosted MCP connector for managing Google Ads, Meta Ads, GA4 and Search Console from chat
- 8 tools
- Remote
- Freemium
- 8.1K GitHub stars
- Marketing and SEO

Tapetide - Indian Stock Market MCP
MCP serverIndian NSE and BSE stock market data for AI assistants through 48 MCP tools
- 17 tools
- Remote
- Freemium
- 150 GitHub stars
- Finance and payments

World Monitor
MCP serverLive geopolitical intelligence dashboard with an API and MCP server for agents
- Remote
- Freemium
- Analytics and data

CompanyData
MCP serverSearch and enrich company records from official registers via the CompanyData API
- 3 tools
- Remote
- Paid
- 3 GitHub stars
- Analytics and data