Mervext

What the Mervext MCP server does

The Mervext MCP server connects MCP-compatible clients to the hosted Mervext document-processing engine at https://api.vstudiodigital.com. It accepts PDFs, images, and text, then returns structured JSON for supported document types. Extraction results include confidence values for individual fields, which lets an agent distinguish well-supported values from uncertain ones.

The server exposes five tools:

• extract_document processes one document using a built-in schema or a caller-provided schema_json.
• classify_document identifies the document type without extracting fields.
• list_schemas returns the available built-in schema identifiers and the data each schema extracts.
• check_balance reports the current plan, remaining credits, and usage.
• delete_document requests explicit deletion of a document by its document ID.

How it works

The Mervext MCP server is a thin wrapper around the deployed Mervext engine. It does not maintain application state, store API keys, or keep client sessions. For hosted use, an MCP client sends requests to https://mcp.vstudiodigital.com/mcp using Streamable HTTP and includes the Mervext key in an X-API-Key or bearer authorization header for each request.

Local clients can run the stdio entry point instead. A local HTTP entry point is also available for testing and listens on port 8787 by default. The hosted deployment runs on Cloudflare Workers and provides GET /healthz for health checks. The documented smoke tests exercise both stdio and hosted HTTP paths against the live engine.

Setup and configuration

Install dependencies and build the project with:

npm install
npm run build

For stdio, set the required MERVEKT_API_KEY variable and start node dist/stdio.js. The same key can be supplied to the hosted HTTP client through request headers. A free trial key is available from Mervext, and the repository also documents a device-key endpoint at /v1/keys/device.

To run the local HTTP transport, use node dist/http.js. The engine URL defaults to https://api.vstudiodigital.com; set MERVEKT_API_URL to override it. Cloudflare Workers deployment uses npx wrangler deploy and the repository's wrangler.toml configuration.

Limitations and notes

Processing consumes credits, and the repository's smoke-test examples note that a run may use one or two credits. The server processes one document per extraction call. Classification does not extract fields, while extraction requires either a built-in schema identifier or a custom JSON schema.

The API key remains the responsibility of the MCP client. The worker stores neither the key nor server-side session state. delete_document provides an explicit deletion operation, but the supplied material does not define retention periods, supported built-in schema names, or the full shape of extraction responses.

The Mervext MCP server is licensed under MIT. Its hosted endpoint is intended for clients that support HTTP and request headers; local stdio is available for clients that use that transport.