AI Assistant Integration

Route Source Parts MCP to custom assistants

Forward Source Parts tools to internal assistants and fine-tuned models while keeping a consistent MCP contract across vendors.

Share a single MCP client

Instantiate the official SDK once and pass the client handle to each assistant. Centralizing credentials reduces the blast radius of a leaked token.

Create a shared MCP client

import { createMcpClient } from "@sourceparts/mcp-sdk";

const client = createMcpClient({
  apiKey: process.env.SOURCEPARTS_API_KEY!,
});

Define a routing layer that translates assistant intents into MCP tool calls. Start with search and quoting, then branch into automation triggers.

Expose intent metadata (confidence scores, conversation IDs) as MCP tool arguments to preserve traceability.
Capture latency metrics per assistant to understand which model families need parallelization or caching.