io.github.Nizoka/pdfnative-mcpfiles

pdfnative MCP — PDF/A & PDF/X-4 generation, PAdES signing & introspection

Verified · 26 days ago

PDF MCP: generate PDF/A & PDF/X-4, sign & verify PAdES (LTV), forms, merge, split, encrypt. 28 tools

Install

claude mcp add pdfnative-mcp -- npx -y [email protected]

Requires: PDFNATIVE_MCP_OUTPUT_DIR, PDFNATIVE_MCP_CACHE_DIR, PDFNATIVE_MCP_PORT, PDFNATIVE_MCP_HTTP_TOKEN, PDFNATIVE_MCP_MAX_INFLATE_BYTES, PDFNATIVE_MCP_CREATION_DATE, SOURCE_DATE_EPOCH, PDFNATIVE_MCP_TSA_URL, PDFNATIVE_MCP_TSA_AUTH, PDFNATIVE_MCP_REVOCATION, PDFNATIVE_MCP_NETWORK_ALLOWED_HOSTS, PDFNATIVE_MCP_NETWORK_TIMEOUT_MS

Our take

A single-author TypeScript MCP server exposing 19 tools over the author's own zero-dependency pdfnative engine: PDF and PDF/A generation from structured blocks, tables, AcroForms, barcodes, 24-script international text, merge/split/extract-pages, PAdES CMS signing and verification, PDF/UA validation, and Factur-X/ZUGFeRD attachment round-trips — an unusually deep spec-compliance surface for a hobby project, with CI, CodeQL, and OpenSSF Scorecard to back it. Two caveats: the registry blurb's 'AI-powered' is marketing fluff — the engine is deterministic PDF plumbing, there is no AI inside — and the popularity numbers are lopsided (1 star, 0 watchers, yet ~807 weekly downloads); the download curve spikes around this project's very frequent releases, consistent with npm mirror and scanner traffic rather than deliberate inflation, and the 2023-vintage owner account with a companion library repo reads as a genuine solo builder, not a bot farm. Text extraction is explicitly best-effort, and the author's refusal to ship a fake overlay-based redact_pdf is a point of honesty. For agents that need to produce signed, standards-compliant PDFs (e-invoices especially) locally without a paid API — just discount the download count and the AI adjective.

reviewed by hand · 2026-07-05

Something wrong or dead here? Report it