What is doc.cheap?
doc.cheap is a document recognition service that reads passports, national ID cards and driver's licences from a photo or scan and returns the printed fields plus the machine-readable zone as JSON — in a single HTTP call. It ships with an OpenAPI 3.1 description, a public sandbox key and an MCP server, so the same recognition works both from your own code and from MCP-capable assistants. It's built for developers wiring document checks into check-in, onboarding and KYC flows.
Who it's for
- Backend and API developers — who want structured document fields back from one HTTP call instead of building OCR and parsing pipelines themselves.
- Teams building check-in or onboarding flows — where a passport or ID scan needs to become usable data quickly.
- Developers working with AI assistants — who want an MCP client to read a document and return separated fields rather than a text blob to re-parse.
Key features
One-call recognition over HTTP
Send a photo or scan and get back the printed fields and the machine-readable zone as JSON. The API is described with an OpenAPI 3.1 specification, and a public sandbox key is available for trying it out.
Structured fields with per-field confidence
Rather than returning a page of text, doc.cheap separates holder details (given names, surname, birth date, sex, nationality) from document details (kind, country, issuing state, number, series, issue and expiry dates, expiry status and days remaining). Each field carries its own confidence, read separately from the machine-readable zone and the printed visual zone.
Machine-readable zone verdict
The MRZ comes back with a verdict — passed, failed or absent — along with the reason and the lines themselves. That lets a caller see whether the two readings agree instead of assuming they do.
MCP server for assistants
The MCP server is published as @doc-cheap/mcp on npm and in the official MCP registry, and runs either over stdio (launched by your client) or over Streamable HTTP. It's a thin client of the public HTTP API and holds no data of its own.
What stands out
Structured fields with confidence scores beat a wall of OCR text — especially when a model is the one reading them.
An OCR tool that returns raw text hands the model something to re-parse, and a model asked to re-parse a date will eventually invent one. doc.cheap returns the fields already separated, with the MRZ and visual-zone readings kept distinct so agreement is visible rather than assumed. Failures come back as one readable line in an error block — never a silent empty result — so the caller can act on it.
Worth checking out if…
You're building check-in, onboarding or KYC flows and want document data as JSON from a single call. It's also a good fit if you're giving an AI assistant document-reading ability and want structured output rather than text to re-parse. With 100 free recognitions every month and no charge for a scan with no readable document, it's low-risk to try.
