EPC Tender Screening MCP

An MCP-only agent tool server for evidence-based EPC tender screening. It exposes HTTP and stdio MCP transports so Claude, ChatGPT, and other MCP-compatible agents can upload company profiles and tender text, extract requirements, match evidence, generate compliance matrices, and produce human-reviewable bid/no-bid memos. The showcase repo includes a screen_tender orchestration tool, deterministic evals, traceable evidence, OAuth-compatible connector routes, CI checks, and SSRF-conscious public source search.

Proof points

  • 10 MCP Tools
  • HTTP + stdio
  • Deterministic Evals

Technologies

TypeScript, MCP, Next.js, Zod, Agent Tools, Evals, CI, OAuth

Links

Problem

EPC teams need faster tender screening, but generic chat workflows often mix real documents with dummy context, lose evidence traceability, and produce recommendations that are hard to review.

Approach

Built an MCP-only TypeScript server with HTTP and stdio transports, structured Zod schemas, local upload/connect tools, requirement extraction, company profile matching, compliance matrix generation, and a screen_tender orchestrator. Added OAuth-compatible routes for custom connector registration, deterministic evals, evidence traces, and guarded public-source search.

Results

Delivered a portfolio-ready MCP project with 10 tools, explicit real-data status checks, no sample fallback, CI quality gates, npm audit passing, and a public GitHub showcase repo ready for Claude, ChatGPT, and other MCP-compatible agents.

EPC Tender Screening MCP: Portfolio Engineering Report

1. Executive Summary

EPC Tender Screening MCP is an agent-agnostic Model Context Protocol server for tender screening. The product is intentionally MCP-only: no dashboard, no broad automation, and no tender auto-submission. It focuses on the smallest credible workflow for human-reviewed EPC bid/no-bid decisions.

2. Core Workflow

  • Connect or upload a company profile
  • Connect or upload a tender brief or tender document text
  • Extract structured tender requirements
  • Match requirements against company capabilities and evidence
  • Generate a compliance matrix with gaps, risks, and next actions
  • Generate a bid/no-bid memo for human review

3. MCP Tooling

  • connect_company_profile_text
  • connect_current_tender_text
  • connect_tender_source_urls
  • get_data_status
  • extract_tender_requirements
  • match_company_profile
  • generate_compliance_matrix
  • generate_bid_no_bid_memo
  • search_historical_proposals
  • search_public_tenders
  • screen_tender

4. Engineering Quality

  • HTTP MCP endpoint for remote connectors and stdio MCP for local clients
  • OAuth-compatible connector routes for Claude.ai-style custom connectors
  • Deterministic eval suite covering extraction, matching, compliance, and full screening
  • Evidence traces and explicit missing-data errors to avoid false-positive screening from dummy data
  • SSRF-conscious public source search with URL allowlisting and private IP blocking
  • GitHub Actions CI for lint, build, evals, and npm audit

5. Impact

This project demonstrates practical MCP product engineering: tool schemas, connector compatibility, production hardening, and evaluable agent workflows. It is designed as a portfolio-ready case study for MCP-first AI agent systems in a real EPC business context.