Connecting Anthropic Claude to Document Pipelines via Native MCP (/api/mcp)

Autonomous AI agents—whether running inside Claude Desktop, Cursor, or custom multi-agent LangChain architectures—excel at high-level reasoning. However, when instructed to “match this supplier invoice to purchase order PO-9912,” agents have historically struggled because document parsing requires multi-modal token geometry, schema coercion, and OCR correction.

To solve this, ParseRelay shipped a Native Model Context Protocol (MCP) Server exposed directly at /api/mcp.

What is the Model Context Protocol?

Developed by Anthropic, MCP is an open standard that allows AI models to securely discover and invoke external tools and resources over JSON-RPC 2.0. Rather than writing custom glue code for every agent framework, any MCP-compatible client can connect to ParseRelay instantly.

Available ParseRelay MCP Tools

Connecting your agent to https://app.parserelay.com/api/mcp provides three primary tools:

1. parse_document: Submits a PDF or image URL, applies an active parser schema (e.g. invoice-v2), and returns structured JSON with confidence ratings and bounding box coordinates.
2. get_extraction_status: Checks asynchronous job state for multi-page documents, returning line-item progress or human review queue status.
3. list_active_schemas: Lists all published parser schemas and their 7 typed primitives, allowing agents to understand valid payload structures before querying.

// Example MCP Tool Invocation via JSON-RPC 2.0
{
	"jsonrpc": "2.0",
	"method": "tools/call",
	"params": {
		"name": "parse_document",
		"arguments": {
			"documentUrl": "https://storage.enterprise.io/inbound/INV-2026-401.pdf",
			"parserSlug": "supplier-invoice",
			"requireReviewThreshold": 0.9
		}
	},
	"id": 1
}

Agent-Assisted Human Review

Because ParseRelay combines autonomous extraction with optimistic locking (CAS) in the human review queue, agents can triage clean documents instantly while routing ambiguous items (such as unlisted vendor names or mismatched PO numbers) directly to human operators.

This hybrid approach allows finance and logistics teams to automate 90%+ of repetitive document workflows while retaining 100% human oversight on edge cases.