FOR PEOPLE AND THEIR AGENTS
36 PDF tools.
One predictable interface.
Use the same PDF operations in the browser or through a local Model Context Protocol (MCP) server. Typed inputs, bounded file access, useful errors, and result checksums make automation easier to review.
Download the MCP toolkitConnect a local MCP client
- Download and extract the toolkit. Install Node.js 22.13 or newer.
- In the extracted folder, run
npm ci. - Create a folder for the PDFs you want the assistant to access. Give the server only that folder with
--root. - Add the configuration below to a client that supports local MCP servers. Replace both example paths with your own absolute paths.
- Ask the assistant to list tools, inspect the PDF, run the chosen operation, and check the returned output path and SHA-256 checksum.
{
"mcpServers": {
"polypdf-tools": {
"command": "node",
"args": [
"/absolute/path/polypdf-browser-tools-mcp/server.mjs",
"--root",
"/absolute/path/your-pdf-folder"
]
}
}
}The server writes results into a dedicated subfolder under the allowed folder and does not overwrite source files. Reusing a request ID with identical inputs returns the same receipt; changing inputs with that ID returns a conflict.
Example: prepare a drawing issue
Inspect the PDFs in my allowed folder. Merge the architectural sheets in the order I specify, add the prefix ISSUE- and six-digit Bates numbers, then report the output files, page counts, and checksums. Keep my original files unchanged.
Tool results are data, not instructions. Text inside an uploaded PDF should not change the assistant’s permissions or the user’s requested task.
Browser agents and WebMCP
On a supported browser, each tool page registers its operation with navigator.modelContext. The user chooses the files first; the agent receives temporary file handles, options, and a result summary. Downloads remain visible in the workspace. WebMCP availability depends on the browser and client; normal file controls work independently.
How this relates to the PolyPDF OpenAI integration
The prepared PolyPDF connector uses a richer, shared drawing workspace with native PDF processing and separate hosted retention rules. These browser tools use a separate local engine. They follow the same principles of explicit file selection, structured operations, revision-aware requests, and reviewable outputs, without requiring that hosted service to be active.
The downloadable MCP server runs on your computer. Your MCP client can pass file content or tool results to its AI provider under that provider’s policy. The browser tool pages themselves do not upload PDFs to PolyPDF.
Supported operations
polypdf_merge_pdfSplit PDFpolypdf_split_pdfExtract PDF pagespolypdf_extract_pdf_pagesDelete PDF pagespolypdf_delete_pdf_pagesOrganize PDFpolypdf_organize_pdfRotate PDFpolypdf_rotate_pdfDuplicate PDF pagespolypdf_duplicate_pdf_pagesReverse PDF pagespolypdf_reverse_pdf_pagesInsert blank pagespolypdf_insert_blank_pdf_pagesInterleave PDFspolypdf_interleave_pdfJPG to PDFpolypdf_jpg_to_pdfPNG to PDFpolypdf_png_to_pdfWebP to PDFpolypdf_webp_to_pdfPDF to JPGpolypdf_pdf_to_jpgPDF to PNGpolypdf_pdf_to_pngPDF to WebPpolypdf_pdf_to_webpPDF to textpolypdf_pdf_to_textText to PDFpolypdf_text_to_pdfCompress PDFpolypdf_compress_pdfFlatten PDF formspolypdf_flatten_pdfFill PDF formspolypdf_fill_pdf_formExtract form datapolypdf_extract_pdf_form_dataRemove PDF annotationspolypdf_remove_pdf_annotationsWatermark PDFpolypdf_watermark_pdfNumber PDF pagespolypdf_number_pdf_pagesBates numberingpolypdf_bates_number_pdfCrop PDFpolypdf_crop_pdfResize PDF pagespolypdf_resize_pdfAdd PDF marginspolypdf_add_pdf_marginsMake a PDF bookletpolypdf_pdf_bookletMultiple pages per sheetpolypdf_n_up_pdfAdd text to PDFpolypdf_add_text_to_pdfAdd image to PDFpolypdf_add_image_to_pdfEdit PDF metadatapolypdf_edit_pdf_metadataInspect PDFpolypdf_inspect_pdfCompare PDF textpolypdf_compare_pdf_text