Open source · Local first

From PDF
to structured
data.

Readable documents for people.
Traceable data for your tools and agents.

Extract text, tables and layout. Keep every result connected to the original source.

Get started Download release
MIT licensedCLI + Web + MCPLocal processing
Technical note.pdfInteractive sample
✓ Input→✓ Parse→✓ Check→✓ Export
Original · Page 1

Charging energy recorded during the reporting period is set out below.

The table preserves the source labels, values and units.

Parsed output Table · Page 1
Linked to the source region
Illustrative content.
Hover or select a highlighted region.
Source stays visible.Structure stays connected.
One document. Useful outputs.
MD Markdown ·</> HTML ·{ } JSON ·W Word optional
Your files stay local ↗
The workflow

A clear path from
document to evidence.

Start in the local workspace, use the command line, or connect an agent. All three use the same parsing pipeline.

01 / INPUT

Add a PDF

Choose a file locally. Keep the original and its identity alongside the results.

02 / PARSE

Preserve structure

Extract paragraphs, tables, page regions and document metadata.

03 / CHECK

Inspect the source

Follow overlays to parsed content. See ambiguous values and review flags in context.

04 / EXPORT

Use the outputs

Read a complete document or pass source-linked structured data to your next step.

More than extracted text

Keep the meaning
in the structure.

Headers, cells, footnotes and source locations belong together. PDF2Data keeps those connections available for inspection.

Explore the outputs
MD

A complete reading

Continuous Markdown and HTML, with page references and retained tables.

{ }

Data with provenance

Structured JSON, raw text, table cells, source hashes and exact locators.

▧

A visual source map

Color-coded regions connect the original page to its parsed content.

W

An editable document

Optional Word export with document roles, formatting and source structure.

Made for people and agents

Your workflow.
The same engine.

Use a local UI for hands-on review, automate with the CLI, or let an MCP-capable agent inspect a document.

01 / WORKSPACE

See what happened.

Upload locally, follow live processing stages and open outputs as they become available.

pdf2data workspace --workspace ./documentsOpen your workspace
02 / MCP

Give your agent
a source-aware tool.

Parse a file, check the run, then read complete pages, tables or scoped evidence.

pdf2data_parsepdf2data_statuspdf2data_read
Connect an agent
03 / CLI

Make it repeatable.

Use explicit commands, reusable packages and versioned outputs in your own pipeline.

pdf2data --helpRead the command reference
Start locally

Small setup.
Useful output.

Node 20+ and uv are required for the npm launcher. Its first run prepares an isolated Python runtime. OCR and AI models are optional.

NPM LAUNCHER / RELEASE CANDIDATEPackage ↗
  1. Install the launcher
    npm install -g @rudraptpsingh/pdf2data@next
  2. Check your environment
    pdf2data doctor
  3. Open your local workspace
    pdf2data workspace --workspace ./documents

    Then open http://127.0.0.1:8771 in your browser.

Prefer Python? Get the wheel or source archive ↗
MCP / LOCAL STDIO

Install the launcher, run pdf2data-mcp --help once, then add this to your host’s MCP configuration. Replace the paths with your own.

{
  "mcpServers": {
    "pdf2data": {
      "command": "pdf2data-mcp",
      "args": [
        "--workspace", "/absolute/workspace",
        "--allow-dir", "/absolute/pdfs"
      ]
    }
  }
}
MCP setup guide
Built in the open

Inspect the source.
Build the next step.

Connect exports to your search, analysis or drafting workflow. Optional DuckDB and NetworkX examples help inspect structure; Cytoscape-compatible data supports relationship views.

Explore on GitHub Integration notes