DreamPDF PDF Engine + MCP Server

Two small Node packages, published separately, usually used together:

Local-only. No API key. No third-party network calls. Every tool runs against a PDF on your own filesystem, over stdio; nothing is uploaded anywhere.

What it can do

• Open a PDF form, read its AcroForm fields, fill them, export the result

• Build a brand-new fillable PDF form from a declarative field list — no source PDF needed

• Extract and search text (fast path), or run full layout analysis (headings/paragraphs/tables/figures) when the optional self-hosted layout_service companion is running

• Read and edit detected text/layout blocks, then re-export

• A redaction-safe reconstruct export profile that rebuilds a PDF from extracted content plus current field values, rather than editing the original page content in place

See pdf_mcp/README.md for the full tool reference, workflow examples, and known limitations.

Related MCP server: dsh-document-editor

Quickstart

npm install -g @dreampdf/pdf-mcp
claude mcp add dreampdf-pdf -- dreampdf-pdf-mcp

pdf_open({ path: "/absolute/path/to/form.pdf" })
pdf_set_fields({ path: "...", fields: { "<field_id>": "value" } })
pdf_export({ path: "...", outPath: "/absolute/path/to/filled.pdf" })

Why this exists

This engine is the headless counterpart to DreamPDF, a mobile-first PDF form editor: the app is for a human tapping through a form on a phone, this is for an agent doing the same thing programmatically. Installing DreamPDF Pro additionally unlocks pdf_get_profile — an agent can then autofill forms from your saved name/contact/signature profile.

Development

This is an npm workspace with two packages:

npm install
npm test --workspaces

pdf_mcp depends on pdf_server via a normal published semver range; there is no build step.

License

Apache-2.0 — see LICENSE. Copyright Canweb.net.