VS Code + pdfmcp: setup guide
Add pdfmcp to VS Code / GitHub Copilot agent mode via mcp.json.
What you need: a free pdfmcp API key — sign in with your email
(no password) and copy the key shown at first login. Then replace pdfmcp_live_YOUR_KEY in the config below.
Where it goes: .vscode/mcp.json in your workspace (or user-level MCP config)
{
"servers": {
"pdfmcp": {
"type": "http",
"url": "https://mcp.pdfmcp.org/mcp",
"headers": {
"X-API-Key": "pdfmcp_live_YOUR_KEY"
}
}
}
}Steps
- Get a free API key from the pdfmcp dashboard and replace pdfmcp_live_YOUR_KEY in the config.
- Save the file, then open the Command Palette → “MCP: List Servers” and start pdfmcp.
- In Copilot Chat agent mode, the pdfmcp tools appear in the tools picker.
- Try: “Use pdfmcp's pdf_to_text on this file and summarise it”.
Troubleshooting
- 401 / unauthorized: the key header is missing or mistyped. The header name is exactly
X-API-Key(Claude Code usesAuthorization: Bearer …— both work). - Tools don't appear: fully restart the client after editing its config; most clients only load MCP servers at startup.
- Server unreachable: check the public status page — and your network's outbound HTTPS to mcp.pdfmcp.org.
Try it with
Compress PDF On-deviceReduce file size with Ghostscript. Presets: light (printer quality), medium (ebook), heavy (screen, smallest).OCR — searchable PDF API · MCPMake a scanned PDF searchable: pages without a text layer are OCR'd with Tesseract and gain an invisible text layer; pages that already have text are passed through and reported as skipped.PDF to Word (DOCX) API · MCPConvert a PDF to an editable Word document. Text-based conversion: paragraphs and headings are reconstructed from the text layer; exact layout, tables and images are not preserved.
More clients: Claude Desktop MCP setup · Claude Code MCP setup · ChatGPT MCP setup · All MCP setup guides · REST API quickstart