Eco Mind MCP Server

Carbon-aware AI development tools for Claude Code. Track energy usage, optimize context windows, and route to efficient models.

Install

claude mcp add eco-mind -- npx eco-mind-mcp

Optional: set a Gemini API key for smarter optimization (summarization, context planning, semantic caching). Without it, tools fall back to local heuristics.

claude mcp add eco-mind -e GEMINI_API_KEY=your_key_here -- npx eco-mind-mcp

Tools

| Tool | Needs Gemini | What It Does | |------|:---:|---| | analyze_session | yes | Real energy/CO2 from Claude Code JSONL logs | | optimize_file | yes | Compare signature extraction vs Gemini summarization | | suggest_context | yes | Recommend which files to include, summarize, or skip | | compare_models | no | Energy comparison across Haiku/Sonnet/Opus | | project_footprint | no | Carbon history timeline across sessions | | explain_impact | yes | Generate shareable impact narratives | | route_query | yes | Recommend cheapest model for a given task | | optimize_prompt | yes | Compress prompts to use fewer tokens | | diff_context | no | Send only file changes, not full contents | | speculative_advisor | no | Analyze Haiku-draft + Sonnet-verify patterns | | semantic_cache | yes | Cache similar queries using embeddings |

How It Works

• Reads Claude Code session logs from ~/.claude/projects/ for real token counts
• Converts tokens to kWh, CO2, and water usage using energy-per-token research data
• Three context reduction methods: signature extraction (~90%), smart context planning (~66-78%), Gemini summarization (~50-70%)

Part of Eco Mind

• Chrome Extension — live carbon widget on Gemini, Claude.ai, ChatGPT
• MCP Server (this) — real tracking + token optimization
• Dashboard — historical analytics via Supabase
• Landing — eco-mind.tech

Built for SFHacks 2026