---
author: OOMOL
author_url: https://oomol.com/about/
datePublished: 2026-11-11
title: Convert PDFs to Markdown with PDFCraft MCP
description: Connect an AI agent to PDFCraft MCP, convert a sample PDF to
  Markdown, check the result, and reuse the text for summaries and document
  workflows.
lang: en
canonical_url: https://oomol.com/blog/pdfcraft-mcp/
markdown_url: https://oomol.com/blog/pdfcraft-mcp.md
---

![A reader holding an open book in soft natural light, with the headline Read beyond the PDF](/blog/pdfcraft-mcp/en-cover.webp)

A report stored as a PDF can be useful source material for an AI agent. Converting it to Markdown gives you a file you can inspect, edit, and reuse when asking for a summary or extracting information.

[PDFCraft](https://pdfcraft.ai/en/) provides a remote MCP service for PDF and EPUB processing. This walkthrough uses a small public PDF to explain the steps from connection to checking the downloaded Markdown. Conversion consumes PDFCraft credits.

## Connect PDFCraft to your agent

Create an API key in [OOMOL Console](https://console.oomol.com/api-key), then follow the [PDFCraft MCP setup guide](https://pdfcraft.ai/en/docs/getting-started/mcp/) to add the server to your agent. PDFCraft has its own MCP endpoint; add it as a separate server alongside any OOMOL App connections you already use.

Before starting, check that your client supports the transport, protocol version, and Bearer authentication headers listed in that guide. Keep the key in the client's protected settings. The client also needs a way to read and upload file bytes: giving a remote server a local filename does not upload the document.

Ask the agent to call `get_capabilities` and explain the available conversion formats. If it cannot discover the tools, resolve the connection before uploading a file or starting a paid task.

## Start with a one-page PDF

Download [PDFCraft's sample PDF](https://pdfcraft.ai/examples/api-quickstart.pdf) and open it. It contains the phrase `Paper to structured text.` that you can look for in the output. This is a small practice input, not a benchmark for scanned books, tables, or equations.

Attach the PDF through your client's file interface, or make it available in a workspace the agent can read. Then ask:

> Use PDFCraft to prepare a Markdown conversion of the attached sample PDF. Check the supported options and my balance, estimate the conversion cost, and show me the estimate before starting processing. Keep the task UUID so we can resume the same task later.

Review the estimate and authorize processing when you are ready. A useful follow-up request is:

> Start the conversion to Markdown. Check progress at intervals until it finishes, then download the complete result into this workspace. Tell me the task UUID and the saved file location. If it fails, show the error before creating another task.

The service's workflow is to create a task, upload the file, launch processing, check progress, and obtain a download link. Tool discovery alone does not confirm that the upload or conversion succeeded. Keep the task UUID until you have saved and checked the output.

## Check the result before asking for a summary

Open the downloaded result. If it is an archive, extract it and inspect the Markdown and any accompanying assets. Ask your agent:

> Open the converted Markdown and check whether it contains “Paper to structured text.” Show the matching passage and compare it with the sample PDF. Report any missing text instead of filling it in.

Finding that phrase is a basic check that the expected input reached the output. For your own documents, compare headings, paragraph order, tables, numbers, and footnotes against the original. A readable summary can still hide a transcription error, especially in a number or equation.

Once you have reviewed the conversion, you can ask the agent to work with the saved text:

> Summarize this Markdown in five points. For each point, include a short supporting quotation and the source heading when available. Keep the original Markdown unchanged.

The agent writes the summary using the conversion result. PDFCraft supplies the document-processing tools. You can keep the Markdown as a reusable source instead of converting the same PDF each time you ask a new question.

## Resume an interrupted task

If the conversation or client stops while processing is running, give the next session the saved task UUID and ask it to check that task. Avoid starting a second conversion just because you lost the first session's progress display.

If a download link expires, request a fresh link for the same task while its files remain available. If an outstanding balance blocks the full download, add credits and request the result again. PDFCraft's [task lifecycle and recovery guide](https://pdfcraft.ai/en/docs/api/lifecycle/) explains task state and recovery; save completed files before they expire.

## EPUB and translation workflows

PDFCraft also supports PDF-to-EPUB conversion and PDF or EPUB translation. For a book you want to read on an e-reader, ask for EPUB output. For translation, ask the agent to check supported languages with `get_capabilities`, confirm the target language, and estimate the cost before processing.

Choose the output for the next task: Markdown for inspecting and reusing text, or EPUB for reading a book. Review a small representative document before committing to a larger conversion.

## Use the text with your work apps

After reviewing the Markdown, you may want to save a summary in a work app or attach it to a project. Set up the relevant app through the [OOMOL Agent quickstart](/docs/agents/), and specify the destination before asking the agent to write anything there. PDFCraft's conversion server and OOMOL's connected Apps are configured separately.

For compression, merging, or splitting PDFs in a browser, open [PDFCraft's online PDF tools](https://pdfcraft.ai/en/tools/). For agent-driven conversion, start with the [PDFCraft MCP setup guide](https://pdfcraft.ai/en/docs/getting-started/mcp/).
