How We Built an MCP Server for GreenSearch AI — Making ESG Data Accessible to Every AI Assistant
We built a custom MCP server that connects GreenSearch AI's ESG document database to Claude, ChatGPT, and Cursor. Now any AI assistant can search sustainability reports directly.
The Challenge
GreenSearch AI had a powerful RAG-based search engine built on FAISS vector search — but it was only accessible through their web UI.
- ESG analysts and investors increasingly used AI assistants (Claude, ChatGPT) for their daily research workflow
- Competitors were starting to integrate with AI platforms, threatening GreenSearch's market position
- The client needed a way to make their data accessible through every major AI assistant without building separate integrations for each platform
- Users had to leave their workflow, open GreenSearch, run a query, and copy results back — a painful context-switching loop
Our Solution
Phase 1 — Architecture & Planning (2 days)
- Audited GreenSearch's existing FastAPI backend and FAISS vector search infrastructure
- Designed the MCP server architecture: which tools to expose, what data to make available, authentication flow
- Defined 6 MCP tools based on the most common user queries:
search_esg_reports— full-text semantic search across ESG documentsget_company_esg_score— retrieve ESG ratings for a specific companycompare_companies— side-by-side ESG comparison of two companiesget_carbon_disclosure— retrieve carbon emission data from CDP reportssearch_regulations— search EU taxonomy and SFDR regulatory documentsget_report_summary— AI-generated summary of a specific ESG report
Phase 2 — Development (5 days)
- Built the MCP server in TypeScript using the official MCP SDK
- Implemented OAuth 2.0 authentication so each user connects with their GreenSearch account
- Added rate limiting (100 requests/hour per user) to protect the underlying API
- Built comprehensive input validation and error handling for each tool
- Implemented response formatting optimized for LLM consumption — structured data with source citations
- Added audit logging for compliance (ESG industry requires full traceability)
Phase 3 — Testing & Integration (3 days)
- Tested with Claude Desktop, ChatGPT (via plugin), Cursor, and Windsurf
- Validated response accuracy against GreenSearch web UI — 99.2% consistency rate
- Load tested with 50 concurrent users — average response time 340ms
- Security audit: penetration testing on OAuth flow, rate limiting verification
- Created documentation and setup guides for end users
Phase 4 — Deployment & Launch (2 days)
- Deployed MCP server on AWS (ECS Fargate) with auto-scaling
- Set up monitoring (CloudWatch + Sentry) and alerting
- Published to MCP server directories for discoverability
- Created onboarding flow: users connect GreenSearch account, get MCP config, start querying in Claude/ChatGPT
Results
6
MCP Tools
Exposed to the AI ecosystem
99.2%
Accuracy
Consistency with web UI results
340ms
Response Time
Average response time
2 Weeks
Delivery
From kickoff to production
- GreenSearch users can now query ESG data directly from Claude, ChatGPT, Gemini, and Cursor
- 40% reduction in context-switching for ESG analysts (based on user feedback)
- New distribution channel: AI-native users discover GreenSearch through MCP server directories
- Platform positioned as AI-native ahead of competitors
Overview
GreenSearch AI is an ESG and sustainability search platform that helps investors, analysts, and compliance teams find relevant information across thousands of sustainability reports, carbon disclosures, and ESG filings. The platform uses RAG (Retrieval-Augmented Generation) with FAISS vector search to deliver accurate, source-cited answers.
The challenge: GreenSearch's powerful search capabilities were locked inside their web app. Users had to leave their workflow, open GreenSearch, run a query, and copy results back. Meanwhile, AI assistants like Claude, ChatGPT, and Cursor were becoming the primary interface for research — but they couldn't access GreenSearch's proprietary ESG data.
The solution: We built a custom MCP (Model Context Protocol) server that exposes GreenSearch's search API to the entire AI ecosystem. Now any user of Claude, ChatGPT, Gemini, or Cursor can query ESG data directly through their AI assistant — without ever leaving their workflow.
Technical Architecture
User in Claude/ChatGPT → MCP Protocol → GreenSearch MCP Server → OAuth verification → GreenSearch API → FAISS Vector Search → ESG Document Database → Formatted response back to user
Key Takeaways
- MCP servers are the new API — if your product isn't in the AI ecosystem, you're invisible
- Building an MCP server on top of an existing API takes 2-3 weeks, not months
- OAuth 2.0 + rate limiting are essential for production MCP deployments
- The ROI is immediate: new distribution channel + better user experience
Want to Make Your Product AI-Native?
We build production-ready MCP servers in 2-4 weeks. Your product in every AI assistant.
Book a Free Consultation → · Learn About MCP → · MCP Server Development →
"We knew MCP was important but didn't have the expertise in-house. IT Flow AI delivered a production-ready MCP server in two weeks that connects our entire ESG database to the AI ecosystem. Our users love being able to query sustainability data directly from Claude without leaving their workflow. This has become a genuine competitive advantage for us."
James Z., Co-founder, GreenSearch AI
ESG / Sustainability / Climate Tech
Want to build something similar?
Book a free 30-minute consultation with Ilya, our founder & CEO, to discuss your AI project.