The self-verifying OCR API.

space-ocr.com · 2026-08-05 · by Long Nguyen ·generic
Where paper ends, data begins. Receipt, invoice, or any document at all — a mountain of paper becomes a database you can search and query, with every cell verifiable right back to its source on the original image. Snap a photo of any receipt, invoice, document, fax, handwritten note, scan, PDF, or free-form layout — the cells fill themselves. OCR, beyond OCR. Use it directly in the browser, automate it via REST API, or install it as a Claude Code plugin — same engine, same result. Any document becomes a queryable row in a familiar sheet: define your columns once and every upload lands as one row with the same schema. Then query the sheet like a database: GET /view supports where=total>=40000, sort=-invoice_date, limit/offset pagination, select=vendor,total projection, and a boxes flag to drop bbox payloads when you don't need them — all server-side, no OCR re-run, no charge. Address sheets by name (/invoices/march) or by uniqueKey; names are unique within a folder. Click any cell to jump to source pixels on the original image and edit inline; per-field bounding boxes on every extracted value make the AI verifiable, not a black box. Drop fifty documents at once and rows stack up automatically. Organize work in Spaces with folders, memos, and full-text search; export sheets to CSV with array-row expansion. Built for teams and developers alike: one API key (spocr_*) on api.space-ocr.com gives you structured field OCR, full MySpace CRUD, async batch upload, polling jobs, and HMAC-signed webhooks for every event (item.created / upload.received / ocr.completed / ocr.failed). Idempotency-Key, per-key rate limiting, OpenAPI 3.1 spec, X-Request-Id everywhere. Multi-script normalization (Japanese / Korean / Chinese / English), auto-rotation. Pay-as-you-go ¥10/image, no charge on failure, free tier without a credit card. Also available as a Claude Code skill — install with /plugin install space-ocr@space-ocr and Claude extracts, stacks, and queries documents in place. Three document endpoints share one verification contract: POST /ocr/fields returns structured named fields, POST /ocr/markdown returns layout-preserving Markdown with per-element boxes, and POST /ocr/text returns reading-order plain text with per-block boxes — every value carrying the vertices it was read from plus text_verified / needs_review flags from the cross-check, so the coordinates come from the OCR pass rather than the language model. The accuracy figures shown on the page were measured on an internal regression corpus of 333 hand-labelled cells (July 2026), not a public benchmark — and the corpus photos themselves are on the page, each opening with the engine-returned vertices drawn on top.

The self-verifying OCR API.

space ocr

LoginStart free

API infrastructure for document data

The self-verifying OCR API.

The pipeline verifies every value it reads. The flags from that check come back with the value, along with the `vertices` it was read from.

Start free — 100 pages/moAPI docs

The console is your playground — drop a file, see the exact response the API returns.

Seafood delivery slip printed on colored paperReceipt with a highlighter stroke across itGrocery slip dense with item codesTwo receipts laid side by side in one shotUS driver's license specimenCarbon-copy slip stack, shot sidewaysDelivery slip shot at the shop counter

How a page becomes verified data

The five steps the pipeline runs on every call.

1. 01

Read the page with OCR

OCR finds every character and where it sits — the only step that produces coordinates. 2. 02

Pass it to a model

The model reads the image and that text together to work out what the page says. It never invents coordinates. 3. 03

Anchor every value

Each value the model returns is anchored to the exact spot on the page where OCR read it. 4. 04

Cross-check the two

Where they disagree, that spot is cropped and read once more — what still fails is flagged: `text_verified` and `needs_review`. 5. 05

Return values with evidence

Each value comes back with the `vertices` it was read from.

Three formats, and a place to keep them

Pick a format when you upload. What it reads stacks up in a folder, and you can query and search it from there.

POST /ocr/fields

Structured fields

Name the fields you want and get them typed, with line items unfolded. Rows stack into a sheet you can query and export as CSV.

POST /ocr/markdown

Markdown

Headings, paragraphs, lists and tables keep their structure — ready to drop into a doc site, a wiki, or an LLM context window.

POST /ocr/text

Plain text

The raw reading, paragraph by paragraph. The plainest way to index, search or diff a scan.

POST /create

Folders

Results stay in a folder you named. Folders nest, and sheets, document bundles and memos all live in them.

GET /view

Query

Pull the stacked rows back with filters. where · sort · select and pagination run on the server.

GET /space

Search

Walk the tree, or type once and everything matches: folder and file names, memo bodies, and the cell values read off the page.

Let an agent file it and fetch it

Connect one MCP server. Hand it photos and it creates the folders and sheets they call for, stacking each photo as a row, then pulls back only the rows you ask about later. The results stay there, so there is no database to stand up.

MCP endpointhttps://mcp.space-ocr.com/mcpSee the MCP setup

  • space\_createCreate folders and sheets
  • space\_uploadUpload photos as rows
  • space\_viewQuery with filters
  • space\_deleteDelete, after confirming

    • Anywhere that speaks MCP

    ClaudeChatGPTPerplexityVS CodeCursor

    Any document into data you can check

    One API call turns a photo into data.

    Item

    Unit price

    Shimaro-Yaka Toast (6 slices)

    ¥98

    Toyama Preference Koshihikari Rice (5kg)

    ¥1,980

    Calpis Soft Drink (800 g)

    ¥698

  • 01
    • Hallucinations get caught in the pipeline Every value is cross-checked against Vision OCR — uncertain ones return `needs_review`.
  • 02
    • Coordinates as proof Each value carries `vertices` pointing into the source image.
  • 03
    • Structured JSON in one call Define fields once, get schema-shaped JSON back.
  • 04
    • Ready the moment you sign up No SDK, just curl. 100 free pages monthly.
  • 05
    • Async + signed webhooks `/jobs` polling · HMAC webhooks · OpenAPI 3.1.
  • 06
    • No database to stand up Create a folder and a sheet, rows pile up, and `/view` pulls them back with filters.

    Get your API keyRead the docs→

    Numbers measured on these photos

    Measured on real field photos, not scans: dim light, tilted angles, uneven quality, no cleanup. Click one to see the vertices the engine returned.

    Phone photo of a delivery slip under a hand shadowOrder sent as a messenger screenshotDelivery slip where one record spans two linesThermal receipt with two-line product recordsStatement with a column of repeated digitsNine-column items table shot from above

    22 / 0cells fixed by the verification stages, versus cells they broke

    92.1%of returned coordinates have the page's own OCR spelling the value inside the box

    6.4×a cell flagged text\_verified: false is 6.4× likelier than average to be a real problem — 3 out of 4 flags are right

    Internal regression corpus of 333 hand-labelled cells from real field photos, cached replay, July 2026. Not a public benchmark. See how it was measured →

    Verification you can see and touch

    The same checked values the API returns, in a screen built to read them: point at a value and the photo answers.

  • 01
    • Drop the photos in No setup, no code. Drag a stack of photos into the browser and the rows fill themselves.
  • 02
    • Point at a value, the photo answers Hover a cell and the exact region of the original lights up — and it works the other way too, from the photo back to the value.
  • 03
    • Doubt is visible, not hidden Values that failed the cross-check stand out in colour instead of sitting quietly wrong, so you only review what deserves it.
  • 04
    • Fix it where you found it Correct a value with the original right beside it. No second window, no hunting for the page it came from.
  • 05
Take it with you Export a sheet as CSV that opens cleanly in Excel, or a doc bundle as .md / .txt.

Try it freeRead the guides→

Frequently asked questions

What kinds of documents can it read?

How is this different from regular OCR?

Can the AI make coordinates up?

What happens when it reads something wrong?

Can it handle line items and tables?

How does it fit into an existing system?

What happens to the original photo?

How much does it cost?

Services running on space ocr

CONNECTiON

space-ocr.com · 14 images · 2 videos · 1 links

Comments (0)

No comments yet