Migrate from PDFShift in one line

The whole migration is two values

Same POST, same JSON body, same X-API-Key header. Their call:

curl -X POST \
  https://api.pdfshift.io/v3/convert/pdf \
  -H "X-API-Key: sk_your_pdfshift_key" \
  -H "Content-Type: application/json" \
  -d '{"source": "https://example.com", "format": "A4", "margin": "10mm", "quality": 90, "timeout": 30}'

This API's:

curl -X POST \
  https://webcapture.io/v3/convert/pdf \
  -H "X-API-Key: wc_your_webcapture_key" \
  -H "Content-Type: application/json" \
  -d '{"source": "https://example.com", "format": "A4", "margin": "10mm", "quality": 90, "timeout": 30}'

The two bodies are byte-identical; only the host and the key move. Run the second call with -o page.pdf, because there is no URL to fetch afterward: the response body is the document itself, with its content type, and x-credits-cost, x-credits-used, and x-credits-remaining headers beside it.

Already holding an HTML string? Keep it in source: {"source": "<html><body><h1>Invoice 42</h1></body></html>"} renders directly. Raw HTML cannot carry http_headers, cookies or auth on this shim; use the native /v1 html and baseUrl fields when relative assets or origin-pinned fetch controls are needed.

The 40 parameters, mapped

Three columns: their parameter, what it becomes on this endpoint, and the status. The rows are generated from the shim's own source, the code that actually decides, so the table cannot promise a mapping that is not performed.

What maps (31)

What refuses (9)

Present in a body, these answer a 422 whose error names every offending parameter at once; nothing is dropped silently. Anything outside both lists is ignored silently, matching pdfshift's own tolerance for unknown fields, so a drop-in caller never hard-stops on a harmless extra.

Timeout, said plainly

pdfshift's timeout counts seconds: their spec allows up to 300, and paid plans wait up to 100 seconds (re-checked 2026-10-03). This endpoint clamps the converted ask to 60,000ms, and the value behaves as a ceiling on the job budget, not a timer you set: a caller asking for timeout: 5 waits up to this endpoint's budget, not five seconds. The ceiling caps the wait; the job budget decides the actual bound.

Clamped values are not surfaced in the response: raw bytes leave no channel for notes, so send timeout at or under 60 and delay at or under 10000 rather than relying on the clamp.

Everything else in the envelope

Errors keep their shape: {"success": false, "error": "...", "code": ...}. A malformed value is a 400; a capability refusal, an unsupported parameter or raw HTML combined with fetch controls, is a 422 naming itself; out of credits is 403 with the quota numbers in an errors object, and a phi request (is_hipaa) from an account without an executed BAA is 403 too; a render past the budget is 408; a failed render is 502. Authentication takes their X-API-Key header, and a Bearer header works too, so one credential can serve both surfaces.

Where this fits

The money side, one render per document against their credit-per-5Mb model, is on webcapture vs PDFShift. The native API this shim fronts, with its own contract and Markdown output, is in the docs; the product page keeps engine, prices, and quick start together.

Method

Checked 2026-10-05. The mapping table is generated from this API's shim source: the exported refusal list and the mapping function beside it, so the columns describe the code that runs, not a spec retype. The PDFShift side of the table comes from pdfshift.io's OpenAPI spec and live docs. Questions or corrections: hello@webcapture.io.