MCP HospiRFQ Processor

mcp_hospirfq_processor exposes RFQ, travel, and hospitality operations as SilvaEngine MCP tools backed by the ai_rfq_engine GraphQL API.

The package provides a processor facade and MCP configuration for a host runtime. It is not a standalone MCP server executable.

Capabilities

The module registers 38 tools across these areas:

• RFQ request creation, updates, item assignment, and quote generation
• Item and provider inventory lookup
• Quote updates, pricing, discounts, and installments
• Availability checks and hold lifecycle management
• Bundle and itinerary discovery
• Cancellation policy lookup
• Hospitality catalog inquiry
• RFQ file and segment-contact operations

Flights use the same tools as other hospitality products. Domain behavior is expressed through item data such as item_type="flight", pricing_mode="per_pax_type", uom="seat", dated batches, and passenger breakdowns.

Architecture

MCPHospiRFQProcessor composes domain mixins over a shared GraphQLBackedProcessor:

MCP host
  -> MCP_CONFIGURATION
  -> MCPHospiRFQProcessor
  -> domain mixins
  -> GraphQLClient
  -> ai_rfq_engine

Each mixin owns one domain concern and uses the shared GraphQL client. The facade initializes that client once and exposes the combined tool surface.

Installation

Python 3.11 or later is required.

python -m pip install -e ".[dev]"

silvaengine-utility must be available from the configured package source.

Configuration

The host creates the processor with a logger and settings dictionary, then assigns the request-specific endpoint and partition identifiers:

import logging

from mcp_hospirfq_processor import MCPHospiRFQProcessor

settings = {
    "graphql_modules": {
        "ai_rfq_engine": {
            "class_name": "AIRFQEngine",
            "endpoint": "https://example.test/graphql/{endpoint_id}",
            "x_api_key": "replace-me",
        }
    },
    "sales_rep_emails": {},
}

processor = MCPHospiRFQProcessor(logging.getLogger(__name__), **settings)
processor.endpoint_id = "endpoint-id"
processor.part_id = "partition-id"

The GraphQL module key must be ai_rfq_engine. The endpoint template may use {endpoint_id}. Requests send x-api-key and Part-Id headers.

Usage

The MCP host loads:

from mcp_hospirfq_processor import MCP_CONFIGURATION, MCPHospiRFQProcessor

Tool arguments use snake_case. The processor converts GraphQL variables to the backend's expected camelCase fields and converts response keys back to snake_case.

Availability and cancellation operations require a partition_key. Holds also require a provider item, service window, and quantity. A hold should be confirmed when a booking is committed, released when abandoned, or expired by the operational expiry process.

Development

Run the local mocked test suite:

python -m pytest -q
python -m compileall -q mcp_hospirfq_processor

The current tests mock GraphQL execution. They verify tool registration, argument mapping, response normalization, error propagation, status transitions, and flight/hospitality conventions. They do not validate a live ai_rfq_engine deployment, network authentication, inventory contention, or scheduled hold expiry.

See docs/DEVELOPMENT_PLAN.md for implementation status, ownership boundaries, and remaining validation work.