E-Rate Tools MCP Server

MCP server providing 9 tools for querying USAC E-Rate datasets via Model Context Protocol.

Remote Server

Production endpoint: https://mbabb.fi.ncsu.edu/e-rate-mcp

Client Configuration

ChatGPT (Direct Connection)

ChatGPT connects directly to the remote /mcp endpoint via JSON-RPC 2.0 protocol.

Setup:

1. Go to: Settings → Apps & Connectors → Advanced settings → Enable Developer Mode
2. Go to: Settings → Connectors → Create
3. Configure:
   • Name: E-Rate Tools
   • Description: USAC E-Rate data API with 9 tools for funding requests, line items, budgets, and entity data
   • Connector URL: https://mbabb.fi.ncsu.edu/e-rate-mcp/mcp

No local installation required.

Claude Desktop (Requires NPM Proxy)

Claude Desktop only supports stdio transport, so a local TypeScript proxy is required to bridge stdio to the remote HTTP server.

Installation:

npm install -g e-rate-mcp

Configuration:

Add to your Claude Desktop config file:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "e-rate-tools": {
      "command": "npx",
      "args": ["-y", "e-rate-mcp"]
    }
  }
}

The proxy connects to https://mbabb.fi.ncsu.edu/e-rate-mcp by default.

Tools

All tools support cursor-based pagination and return inline data (NO resource URIs).

Pattern: Returns 50 rows per page by default, use next_cursor to fetch additional pages.

1. usac_download - Download datasets with pagination (page_size: 50, max: 100)
2. usac_stream - Stream large datasets in chunks
3. usac_dictionary - Get dataset schema and column information
4. usac_get_frn_status - FRN status with auto-deduplication + pagination
5. usac_get_line_items - Line items with auto-deduplication + pagination
6. usac_get_basic_info - Basic application info with auto-deduplication + pagination
7. usac_get_recipients - Recipients of service + pagination
8. usac_get_cat2_budgets - Category 2 budgets (FY2021+) + pagination
9. usac_get_entity_data - Supplemental entity data + pagination

Pagination Example:

// First page
{"view": "CONSULTANTS", "year": 2025, "page_size": 50}

// Returns:
{
  "total_rows": 975,
  "returned_rows": 50,
  "next_cursor": "NTA=",
  "has_more": true,
  "data": [...]
}

// Next page
{"view": "CONSULTANTS", "year": 2025, "page_size": 50, "cursor": "NTA="}

Datasets

• FRN_STATUS - Funding request status
• FRN_LINE_ITEMS - Service line items
• FRN_RECIPIENTS_OF_SERVICE - Entity-to-FRN mappings
• FRN_BASIC_INFORMATION - Form 471 metadata
• SUPPLEMENTAL_ENTITY_DATA - Organization profiles
• CONSULTANTS - Consultant engagements
• CAT2_BUDGETS - Category 2 budgets (FY2021+)
• RHC_COMMITMENTS_AND_DISBURSEMENTS - Rural Health Care

Development

Local Testing (Stdio)

cd mcp_server
export PYTHONPATH=/path/to/project/src
uv run python server.py

Local Testing (HTTP)

cd mcp_server
export PYTHONPATH=/path/to/project/src
uv run python http_server.py

Server starts on http://localhost:8002

Endpoints:

• POST /mcp - JSON-RPC 2.0 endpoint (ChatGPT)
• GET /resources - List resources
• GET /resources/{uri} - Read resource
• GET /tools - List tools
• POST /tools/{name} - Call tool
• GET /prompts - List prompts
• POST /prompts/{name} - Get prompt
• GET /health - Health check

Deployment

Production deployment via Docker:

./deploy.sh

Builds Docker image and deploys to production server.

Architecture

mcp_server/
├── server/              # Python MCP server
│   ├── main.py          # Server orchestration
│   ├── http_wrapper.py  # HTTP + JSON-RPC 2.0 wrapper
│   ├── registry.py      # Tool specifications
│   ├── tools.py         # Tool handlers
│   ├── resources.py     # Resource handlers
│   ├── prompts.py       # Prompt templates
│   ├── cache.py         # Download cache
│   └── descriptions.py  # Dataset metadata
├── src/
│   └── index.ts         # NPM proxy (stdio → HTTP)
├── http_server.py       # HTTP entry point
├── server.py            # Stdio entry point
├── Dockerfile           # Docker image
└── package.json         # NPM package config

NPM Package

The NPM package e-rate-mcp is a thin TypeScript proxy that:

1. Accepts stdio requests from Claude Desktop
2. Forwards to remote HTTP server at https://mbabb.fi.ncsu.edu/e-rate-mcp
3. Returns responses via stdio

Source: src/index.ts (200 lines)

Why needed: Claude Desktop only supports stdio transport, not HTTP.

License

MIT - Part of e-rate-tools project for NC State Connectivity Initiative