TenderAI — MCP Server for Tender & Proposal Management

A production-ready Model Context Protocol server that automates government/enterprise tender workflows: RFP parsing, technical proposal writing, financial proposal assembly, partner coordination, and compliance tracking.

Features

• 18 MCP Tools across 5 domains: Document Intelligence, Technical Proposals, Financial Proposals, Partner Coordination, Past Proposal Indexing & Search
• Hybrid Search: FTS5 full-text keyword search + sqlite-vec vector similarity search with Reciprocal Rank Fusion (RRF)
• 5 Resource URI schemes for knowledge base access: past proposals, templates, vendors, company profile, standards
• 4 Workflow Prompts for end-to-end orchestration: tender analysis, executive summaries, partner checks, full proposal workflow
• AI-Powered: Uses Claude to parse RFPs, generate proposal sections, and produce compliance narratives
• Voyage AI Embeddings (optional): Semantic search over past proposals — finds similar projects even without exact keyword matches
• Document Generation: Professional DOCX proposals and XLSX BOM spreadsheets
• SQLite Database: Tracks RFPs, proposals, vendors, BOM items, partners, deliverables, and indexed past proposals
• OAuth 2.0: Built-in OAuth support for claude.ai integration (Dynamic Client Registration + PKCE, auto-approve)

Quick Start

Local Development (stdio)

# Clone and setup
cd tenders
python -m venv venv && source venv/bin/activate
pip install -r requirements.txt

# Configure
cp .env.example .env
# Edit .env — set ANTHROPIC_API_KEY

# Run
python -m app.server

Claude Desktop / Claude Code Configuration

stdio (local):

{
  "mcpServers": {
    "tenderai": {
      "command": "python",
      "args": ["-m", "app.server"],
      "cwd": "/path/to/tenders",
      "env": {
        "ANTHROPIC_API_KEY": "sk-ant-..."
      }
    }
  }
}

HTTP (remote — Claude Code/Desktop):

{
  "mcpServers": {
    "tenderai": {
      "type": "http",
      "url": "https://tender.yfi.ae/mcp",
      "headers": {
        "Authorization": "Bearer <MCP_API_KEY>"
      }
    }
  }
}

Claude.ai (OAuth 2.0):

Set OAUTH_ISSUER_URL=https://tender.yfi.ae in .env, then add https://tender.yfi.ae/mcp as an integration on claude.ai. OAuth flow completes automatically.

Production Deployment

sudo ./setup.sh tender.yfi.ae
# Edit /opt/tenderai/.env — set ANTHROPIC_API_KEY
sudo systemctl start tenderai

Tools

Document Intelligence

Technical Proposals

Financial Proposals

Partner Coordination

Past Proposal Indexing & Search

Resources

Prompts

Knowledge Base

Populate these directories to improve AI-generated content:

data/
├── knowledge_base/
│   ├── company_profile/
│   │   └── profile.md          # Company description, capabilities, differentiators
│   ├── templates/
│   │   ├── executive_summary.md
│   │   ├── technical_approach.md
│   │   └── ...                 # Section-specific templates
│   └── standards/
│       ├── iso27001.md
│       └── ...                 # Standards reference docs
├── past_proposals/
│   ├── tra_network_2024/
│   │   ├── Technical_Proposal.pdf      # Your original submission files
│   │   ├── Cost_Sheet.xlsx             # Financial data for pricing reference
│   │   └── _summary.md                 # Auto-generated by index_past_proposal
│   └── omantel_5g_2024/
│       └── ...
├── rfp_documents/              # Auto-populated by parse_tender_rfp
├── vendor_quotes/              # Vendor quote files
└── generated_proposals/        # Auto-populated output

Search Architecture

Past proposals can be indexed for fast retrieval:

Upload files → index_past_proposal → AI extracts metadata → stored in SQLite
                                                           ├── FTS5 (keyword search, always on)
                                                           └── sqlite-vec (vector search, optional)

• FTS5: BM25 keyword ranking with porter stemming — sub-millisecond search
• Vector: Voyage AI embeddings (512-dim) stored in sqlite-vec — semantic similarity
• Hybrid: Both combined via Reciprocal Rank Fusion (RRF) for best results
• Set VOYAGE_API_KEY in .env to enable vector search (200M free tokens from Voyage AI)

Backup

# Manual backup
./backup.sh /backups/tenderai 30

# Cron (daily at 2 AM)
0 2 * * * /opt/tenderai/backup.sh /backups/tenderai 30

Architecture

app/
├── server.py              # Entry point — FastMCP init and wiring
├── config.py              # Settings from .env
├── tools/
│   ├── document.py        # 4 document intelligence tools
│   ├── technical.py       # 4 technical proposal tools
│   ├── financial.py       # 4 financial proposal tools
│   ├── partners.py        # 3 partner coordination tools
│   └── indexing.py        # 3 past proposal indexing & search tools
├── resources/
│   └── knowledge.py       # 5 resource URI handlers
├── prompts/
│   └── workflows.py       # 4 workflow prompts
├── db/
│   ├── schema.sql         # SQLite schema (10 tables + FTS5 + triggers)
│   ├── database.py        # Async database layer + sqlite-vec + OAuth CRUD
│   └── models.py          # Pydantic models
├── services/
│   ├── llm.py             # Anthropic SDK wrapper (15 prompt templates)
│   ├── parser.py          # PDF/DOCX/XLSX parser
│   ├── embeddings.py      # Voyage AI embedding service (optional)
│   └── docwriter.py       # DOCX/XLSX generator
└── middleware/
    ├── auth.py            # ASGI Bearer token auth (Claude Code/Desktop)
    └── oauth.py           # OAuth 2.0 provider (claude.ai)