Convert Any File to Markdown in Claude & Cursor via MCP

If you've ever watched an AI agent fail because it couldn't "read" a PDF, you already understand the problem. LLMs are phenomenal at reasoning over text. They fall flat when the information is locked inside a binary file. The standard fix — manually converting documents and pasting them in — doesn't scale, breaks agentic workflows, and introduces human error.

The Markdown Anything MCP Server closes this gap entirely. It connects your AI assistant directly to a battle-tested document conversion engine, so Claude, Cursor, Windsurf, or any MCP-compatible client can convert PDFs, Word docs, spreadsheets, images, audio, and 25+ other formats into clean, structured Markdown — mid-conversation, without you lifting a finger.

What Is the Model Context Protocol (MCP)?

The Model Context Protocol is an open standard that lets AI assistants discover and invoke external tools in a structured, reliable way. Instead of prompt-engineering your way around an LLM's file-blindness, you expose a capability through an MCP server, and the AI calls it natively — just like it would use a calculator or a web search tool.

Markdown Anything ships a production-ready MCP server at:

https://markdownanything.com/mcp/conversions

Any MCP-compatible client can discover its tools automatically. No custom wrappers, no brittle function calls.

What Can the Markdown Anything MCP Server Do?

The server exposes four tools that cover the full document-conversion lifecycle:

Tool	What It Does
convert-file	Converts a base64-encoded file into clean Markdown
get-conversion	Retrieves the status and result of a past conversion
check-credits	Returns your workspace credit balance and plan details
list-supported-formats	Lists all supported file types grouped by category

Supported File Formats

The conversion engine handles 25+ formats across every major category:

• Documents: PDF, DOCX, DOC, RTF, ODT, EPUB, TXT, MD
• Spreadsheets: XLSX, XLS, CSV, ODS
• Presentations: PPTX, PPT, ODP
• Web: HTML, HTM, XML, JSON
• Images: JPG, JPEG, PNG (with AI-powered OCR)
• Audio: WAV, MP3, M4A (with speaker detection)

Maximum file size is 10 MB on Starter, 15 MB on Professional, and 50 MB on Business.

Setting Up the Markdown Anything MCP Server

Setup takes under two minutes. Get your API token from Settings → API Tokens in your Markdown Anything workspace dashboard, then add the following to the config file for your client.

Claude Desktop

Edit claude_desktop_config.json:

{
  "mcpServers": {
    "markdown-anything": {
      "type": "streamable-http",
      "url": "https://markdownanything.com/mcp/conversions",
      "headers": {
        "Authorization": "Bearer mda_your_token_here"
      }
    }
  }
}

Claude Code

Add to your project's .mcp.json:

{
  "mcpServers": {
    "markdown-anything": {
      "type": "streamable-http",
      "url": "https://markdownanything.com/mcp/conversions",
      "headers": {
        "Authorization": "Bearer mda_your_token_here"
      }
    }
  }
}

Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "markdown-anything": {
      "type": "streamable-http",
      "url": "https://markdownanything.com/mcp/conversions",
      "headers": {
        "Authorization": "Bearer mda_your_token_here"
      }
    }
  }
}

Replace mda_your_token_here with your real API token. That's it — the client discovers the tools automatically on next launch.

How It Works in Practice

Once configured, you interact with Markdown Anything entirely through natural language. Your AI assistant handles the conversion mechanics behind the scenes.

Example prompts:

• "Convert this PDF to Markdown" → the assistant reads the file, base64-encodes it, and calls convert-file
• "How many credits do I have left?" → calls check-credits and summarizes the response
• "What file types can you convert?" → calls list-supported-formats
• "Show me the result of conversion abc-123" → calls get-conversion with the ID

The convert-file tool accepts three optional flags that unlock advanced capabilities:

• use_enhanced_ai — runs documents through higher-accuracy AI models, ideal for scanned PDFs, low-quality images, and complex layouts
• include_metadata — appends document metadata (author, creation date, page count) to the Markdown output
• optimize_tokens — trims whitespace and redundant structure to reduce token consumption in downstream LLM calls

A typical response looks like this:

{
  "conversion_id": "9e5fcd8a-1b2c-3d4e-5f6a-7b8c9d0e1f2a",
  "markdown": "# My Document\n\nConverted content here...",
  "credits_used": 1,
  "credits_remaining": 499
}

If a conversion fails for any reason, credits are automatically refunded to your workspace.

Why This Matters for RAG Pipelines and Agentic Workflows

Document conversion is rarely the goal — it's the bottleneck blocking the goal. Here's where the Markdown Anything MCP server changes the architecture of real-world AI systems:

RAG pipelines: Agents can pull in source documents at query time, convert them on-the-fly, and feed clean Markdown directly into the retrieval context — no pre-processing step, no stale document store.

Code assistants: A developer can point Cursor at a 50-page API reference PDF and ask questions about it. The agent converts it, reads it, and answers without leaving the editor.

Research agents: Claude can ingest a batch of DOCX reports, spreadsheets, or EPUB books, convert them sequentially, and synthesize findings across all of them in one session.

Autonomous agents: Any agent with MCP support can call list-supported-formats to introspect its own capabilities before deciding whether to attempt a conversion — enabling self-directed, tool-aware behavior.

The key insight is that document conversion becomes a first-class tool the agent controls, not a pre-processing step a human manages.

Credit Costs

The MCP server uses the same credit system as the REST API — no separate pricing tier.

Conversion Type	Cost
Standard	1 credit
Enhanced AI (Starter plan)	3 credits
Enhanced AI (Professional / Business)	2 credits

All plans include 25 free credits per month to start experimenting.

Authentication and Security

The MCP server uses Bearer token authentication scoped to your workspace. All credits, conversions, and rate limits apply at the workspace level — the same as the REST API. Tokens are created and revoked from the workspace dashboard, giving you full control over access.

Who Should Use This?

The Markdown Anything MCP server is built for:

• AI engineers wiring up RAG pipelines who need reliable, structured document ingestion
• Developers using Claude Code or Cursor who work with technical documentation in non-text formats
• Automation builders who want Claude or another agent to process uploaded files as part of a broader workflow
• Researchers who need to analyze large batches of PDFs, reports, or slide decks inside a single AI session

If your work involves feeding documents to an LLM — and you want the AI to handle that step itself — this integration is the most direct path there.

Get Started

1. Create a free Markdown Anything account — 25 credits included, no credit card required
2. Generate an API token from Settings → API Tokens
3. Add the MCP server config to Claude Desktop, Claude Code, or Cursor
4. Ask your assistant to convert a file

Full documentation is available at markdownanything.com/docs/integrations/mcp.

---

Markdown Anything is a document-to-Markdown API built for LLMs and RAG pipelines — part of the Climactic product suite.