ParseRelay

Name: ParseRelay
Author: parserelay

Local

parserelay.appProductivity0Updated 19w ago

Repository Visit website

About ParseRelay

ParseRelay provides a single MCP `scan` tool for turning documents into structured fields with per-field confidence scores. It is for MCP hosts that need synchronous document parsing, with optional dry runs to preview paid rescue credits.

Key features

01Structured fields from document images
02Per-field confidence and review indicators
03Selectable schema, document type, and engine
04Dry-run estimates for paid rescue credits

Use cases

01Extract fields from a document image
02Check which parsed fields need review
03Estimate credits before a paid scan

MCP support

Provides one synchronous `scan` tool to MCP hosts. Results include fields, confidence, review status, source, and metadata; it supports stdio and streamable HTTP transports.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

mcp.json · Universal config

{
  "mcpServers": {
    "parserelay": {
      "command": "npx",
      "args": [
        "-y",
        "@parserelay/mcp"
      ],
      "env": {
        "PARSERELAY_API_KEY": "<PARSERELAY_API_KEY>",
        "PARSERELAY_BASE_URL": "<PARSERELAY_BASE_URL>"
      }
    }
  }
}

Run commandnpx -y @parserelay/mcp

Environment variables to set

PARSERELAY_API_KEYSecretParseRelay API key (prl_live_…). Required for stdio.

PARSERELAY_BASE_URLAPI base URL. Defaults to https://api.parserelay.app; point at a local worker for testing.

Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.

Packages

npm@parserelay/mcp0.1.1stdio

Tools

Saved tool list · updated 3w ago

Parameter	Type	Required	Description
ocr	any	No	OCR backend selection. Use backend `passthrough` with `text` to skip OCR.
image	string	Yes	The document to scan: an https URL or a base64 data URI (data:image/...;base64,...).
model	string\|null	No	Pin a specific model, e.g. "claude-haiku-4-5".
engine	enum	No	Extraction mode. Default `auto` (deterministic-first, LLM rescues only flagged fields).auto \| ocr \| ocr+rescue \| ocr+check \| vision
schema	any	No	What to extract, in ONE of two forms: a field-name array, OR a JSON Schema object with a `properties` map. Do NOT pass a flat {"field":"type"} object. Omit for a generic document.
dry_run	boolean	No	Preview only: returns which fields WOULD trigger a paid rescue + estimated credits, spends nothing.
doc_type	enum	No	Optional document-type hint.receipt \| invoice \| id \| business_card \| freeform
model_key	string\|null	No	Bring your own provider key — you're billed plumbing only; the model bill is on your key.

Details

From the project README

@parserelay/mcp

ParseRelay's scan operation as an MCP tool — so any MCP host (Claude Desktop, Cursor, …) can parse a document into structured, confidence-scored fields. Same contract as the REST API and the <DeadSimpleMicroScanner> component; one tool, scan.

The `scan` tool

Input mirrors the sync ScanRequest — image (required), plus schema, doc_type, engine, ocr, dry_run, model, model_key. (relay webhooks are omitted: a tool call is synchronous, so you get the envelope back inline.) The result is the full ScanEnvelope as JSON: fields, per-field confidence, needs_review, field_source, and meta (engine, credits, tokens).

Set dry_run: true to preview which fields would trigger a paid rescue — and the estimated credits — without spending anything.

Run it

stdio (Claude Desktop, Cursor)

{
  "mcpServers": {
    "parserelay": {
      "command": "npx",
      "args": ["-y", "@parserelay/mcp"],
      "env": {
        "PARSERELAY_API_KEY": "your-key",
        "PARSERELAY_BASE_URL": "https://api.parserelay.app" // optional
      }
    }
  }
}

streamable-HTTP (remote / hosted)

PARSERELAY_API_KEY=your-key PORT=8080 node node_modules/@parserelay/mcp/dist/http.js
# → POST http://localhost:8080/mcp

Stateless: a fresh server per request, so it scales horizontally. Auth per request via Authorization: Bearer <key>, falling back to PARSERELAY_API_KEY for single-tenant setups.

Programmatic

import { createMcpServer } from "@parserelay/mcp";
import { ParseRelayClient } from "@parserelay/client";

const server = createMcpServer(new ParseRelayClient({ apiKey }));
// then wire your own transport: await server.connect(transport)

Config

Env var	Required	Default	Notes
`PARSERELAY_API_KEY`	stdio: yes	—	HTTP can override per request via `Authorization: Bearer`.
`PARSERELAY_BASE_URL`	no	`https://api.parserelay.app`	Point at a local worker for testing.
`PORT`	no	`8080`	HTTP transport only.

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

Read the getting-started guide

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.