pdf-mcp

Let your AI agent pull text and tables out of PDFs. An MCP server for invoices, reports, and statements, where the data lives in tables the model can't read from a pasted blob.

When you paste a PDF into a prompt, the columns collapse and the table turns to mush, so the model guesses at the numbers. This extracts the actual table structure with deterministic code, so the agent gets clean rows and never invents a cell.

What it turns a PDF into

A PDF invoice table like this:

Item     Qty   Price
Widget    3    12.50
Gadget    1    40.00
Bolt     10     0.25

comes back as structured rows (or CSV), not a flattened line of text:

[["Item","Qty","Price"],["Widget","3","12.50"],["Gadget","1","40.00"],["Bolt","10","0.25"]]

Related MCP server: pdfmux

The tools it gives an agent

Quickstart

pip install "pdf-agent-mcp[mcp]"

Add it to your MCP client (e.g. Claude Desktop):

{
  "mcpServers": {
    "pdf": { "command": "pdf-agent-mcp" }
  }
}

Now your agent can answer "pull the line items out of this invoice" by reading the PDF, not guessing.

Also usable from plain Python

from pdf_mcp import extractor

extractor.extract_tables("invoice.pdf")      # {'tables': [{'rows': [...]}], ...}
extractor.table_to_csv("invoice.pdf")        # clean CSV of the first table
extractor.extract_text("report.pdf", page=1)

Tests

python -m unittest discover -s tests     # builds its own test PDF, runs anywhere

License

MIT