96.8%
less document data sent to the model
A 792 KB whole-PDF read became 25 KB of retrieved pages.
Agent-first PDF and Markdown Reader
MarkPDF
An open-source, minimal PDF and Markdown reader with local-first annotation, OCR, visual signing, forms, image-to-PDF conversion, and AI-powered semantic search. Expect 30-80% reduction in input token cost on large documents.
Benchmarked against whole-PDF reading
Ask a question about a PDF without paying for the whole PDF.
MarkPDF ships an MCP server, so a coding agent can search your local index and read only the pages that matter. On a short technical report, that route matched the accuracy of loading the entire PDF into Claude while sending a fraction of the document.
84.1%
fewer fresh input tokens
199,250 tokens fell to 31,670 on the same question.
79.1%
lower cost per answer
The same 100-word answer cost $0.44 instead of $2.09.
Same
answer accuracy
Both routes answered correctly from the source passage.
Documents too large to read at all
On a reference book of more than 600 pages, reading the file whole produced a 15 MB payload that the model API rejected outright, so the agent never answered. MarkPDF answered the same question from an 18 KB excerpt, transferring 99.9% less of the document.
A skill that keeps retrieval narrow
The bundled markpdf-retrieval skill drops broad outline calls and redundant searches. Against an agent choosing its own MarkPDF route, it cut the serialized tool payload by 61.8% and 81.3% on the two documents, and cost by 29.6% and 39.2%.
Built for local document work
More than a basic PDF viewer.
MarkPDF brings reader, editor, converter, and AI-search workflows into one local desktop surface.
Minimal PDF and Markdown reading
Open PDFs and Markdown files in a standalone desktop app without the clutter of heavier PDF suites.
Local semantic search
Index document text locally with ONNX embedding models and search related passages without sending PDF text to an AI provider.
AI provider and CLI-agent ready
Detect local CLI agents and connect OpenRouter, Ollama, LM Studio, Anthropic, or OpenAI-compatible endpoints.
OCR and image-to-PDF conversion
Use OCR fallback for scanned pages and convert groups of images into PDFs from the desktop app.
Annotations, forms, and page editing
Add text overlays, highlights, comments, reorder pages, fill forms, and save standard PDF annotations.
Visual signing
Draw, type, or upload a signature and place it on the page. Certificate-backed digital signatures are not included yet.