PDF MCP server for Claude, ChatGPT, Cursor and VS Code

Give your AI assistant 26 PDF Vector tools: parse PDFs and scans to Markdown, pull bank statement, invoice and ID fields as JSON, and search academic papers and grants. It is a hosted server, so you paste one URL and sign in.

https://global.pdfvector.com/mcp

Set up your client Get a free API key

MCP tools: 26 tools
Transport: Streamable HTTP, hosted
Sign-in: OAuth or API key
Free to start: 100 credits

All 26 MCP tools

Tool names as your client lists them. Each runs the same API as the web app and bills the connected workspace.

Any document

document_parse
Turn a PDF, Office file, scan or image into Markdown.
document_ask
Answer a question about a document.
document_extract
Return the fields of your JSON Schema as JSON.

Bank statements

bankStatement_parse
Read a statement into text and financial fields.
bankStatement_ask
Answer questions about balances and transactions.
bankStatement_extract
Return statement data matching your schema.

Invoices and receipts

invoice_parse
Read an invoice into text and invoice fields.
invoice_ask
Answer questions about totals, dates and terms.
invoice_extract
Return invoice data matching your schema.

ID documents

identity_parse
Read a passport, license or ID card into fields.
identity_ask
Answer questions about an ID document.
identity_extract
Return ID data matching your schema.

Academic research

academic_search
Search Semantic Scholar, arXiv, PubMed, OpenAlex and more.
academic_fetch
Look up papers by DOI, PubMed ID, arXiv ID or URL.
academic_parse
Find a paper's public PDF and return its full text.
academic_findCitations
Find papers that support a paragraph.
academic_paperGraph
List the papers a paper cites and is cited by.
academic_similarPapers
Find related papers through the citation network.
academic_searchGrants
Search Grants.gov, NIH RePORTER, CORDIS and UKRI.

Free tools

free_bankStatementParse
Parse a bank statement within the free-tier limits.
free_documentParse
Convert a document to markdown within the free-tier limits.
free_generateSchema
Write an extraction JSON Schema from a plain instruction.
free_publicationPdfUrl
Find PDF links for up to 20 papers.
free_rateLimitStatus
Show the remaining free-tier allowance.

Account

account_status
Show the workspace, plan and credit balance (no credits used).
account_checkout
Create a Stripe link to upgrade to Basic or Pro (OAuth only).

Two ways to connect

Both endpoints expose the same tools and use the credits of one workspace. Upgrades always happen in Stripe Checkout, which you open and confirm yourself.

OAuth sign-in

https://global.pdfvector.com/mcp

Sign in with your PDF Vector account and pick a workspace; no key to copy.
Clients discover the sign-in server automatically and register themselves.
Required for account_checkout, which creates a Stripe link to upgrade.

Workspace API key

https://global.pdfvector.com/mcp/api-key

Send Authorization: Bearer <API key> with every request.
Best for headless agents, CI and clients without OAuth.
account_status works; payment links need OAuth.

How to connect PDF Vector's MCP server

Claude and Claude Desktop

Use the one-click link, or open the Connectors settings in Claude and choose Add custom connector.
Name it PDF Vector and paste https://global.pdfvector.com/mcp as the URL.
Click Connect, sign in to PDF Vector and pick the workspace whose credits Claude should use. The connector is then available in Claude on the web and in Claude Desktop.

Add to Claude

Claude Code

Add the server, then run /mcp inside Claude Code and choose PDF Vector to sign in.
To use an API key instead of signing in, add the API-key endpoint with an Authorization header. Claude Code reads PDFVECTOR_API_KEY from your environment when it starts.

Terminal

claude mcp add --transport http pdfvector https://global.pdfvector.com/mcp

# or, with a workspace API key
claude mcp add --transport http pdfvector https://global.pdfvector.com/mcp/api-key \
  --header 'Authorization: Bearer ${PDFVECTOR_API_KEY}'

Cursor

Use the one-click install, or add the server to ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (one project).
When Cursor shows that PDF Vector needs a login, click it and sign in to PDF Vector.

Install in Cursor

~/.cursor/mcp.json

{
  "mcpServers": {
    "pdfvector": {
      "url": "https://global.pdfvector.com/mcp"
    }
  }
}

VS Code and GitHub Copilot

Use the one-click install, or add the server to .vscode/mcp.json in your workspace.
Start the server from the file or the MCP view, sign in when VS Code asks, then use the tools in Copilot Chat agent mode.

Install in VS Code

.vscode/mcp.json

{
  "servers": {
    "pdfvector": {
      "type": "http",
      "url": "https://global.pdfvector.com/mcp"
    }
  }
}

ChatGPT

Custom MCP connectors in ChatGPT need developer mode, which OpenAI offers on selected plans. Turn it on in Settings under Apps and Connectors.
Create a connector, paste https://global.pdfvector.com/mcp as the MCP server URL and choose OAuth.
Sign in to PDF Vector when ChatGPT opens the login page, then enable the connector in a chat.

Other MCP clients

PDF Vector is a remote MCP server over Streamable HTTP. Any client that supports remote servers can connect: Windsurf, Goose, LM Studio, Gemini, Cherry Studio and others.
With OAuth, give the client https://global.pdfvector.com/mcp. It finds the sign-in server through the standard OAuth protected-resource metadata and registers itself.
With an API key, use https://global.pdfvector.com/mcp/api-key and send the header Authorization: Bearer <your API key>.

Generic remote server config

{
  "pdfvector": {
    "type": "http",
    "url": "https://global.pdfvector.com/mcp/api-key",
    "headers": {
      "Authorization": "Bearer YOUR_PDFVECTOR_API_KEY"
    }
  }
}

Things to ask once it is connected

Your assistant picks the tool. File paths work in coding agents that can read your disk.

“Parse ./statements/march.pdf and list every transaction with date, description, amount and balance.”
bankStatement_extract
“Read the invoices in ./inbox and give me a CSV with vendor, invoice number, total and due date.”
invoice_extract
“Is the passport in ./kyc/scan.jpg expired, and does the holder's name match Jane Doe?”
identity_ask
“Convert ./docs/spec.docx to Markdown so I can add it to the repo.”
document_parse
“Find papers from 2024 onward on retrieval-augmented generation for clinical notes and list the ten most cited with DOIs.”
academic_search
“Find citations that support this paragraph, then show which papers cite the strongest one.”
academic_findCitations, academic_paperGraph
“Search NIH RePORTER and Grants.gov for open grants on antimicrobial resistance.”
academic_searchGrants
“How many PDF Vector credits does my workspace have left?”
account_status

Frequently asked questions

What is the PDF Vector MCP server URL?

https://global.pdfvector.com/mcp for OAuth sign-in, or https://global.pdfvector.com/mcp/api-key with an Authorization: Bearer header carrying a workspace API key. Both use the Streamable HTTP transport; nothing needs to be installed or run locally.

Should I connect with OAuth or an API key?

OAuth is easiest in chat apps and editors: you sign in once and pick the workspace, and only OAuth can create upgrade links with account_checkout. An API key suits headless agents, CI and clients without OAuth support; it can read the account status but not create payment links.

How do I give the server a PDF?

Document tools take the file content as base64, so the AI client reads the file and sends it. Coding agents such as Claude Code, Cursor and VS Code do this from files on disk, which makes them the most reliable choice for documents. Academic tools only need a query, DOI or URL and work in any client.

Can it read scanned PDFs and images?

Yes. Scans and photos are read with OCR. Images need the Auto, Pro or Max model; Auto picks a model that can read the file.

How much does it cost?

Tool calls use the credits of the connected workspace, the same as the API: Nano 1 credit per page, Mini 2, Pro 4, Max 8, and Auto bills the tier it selects. account_status uses no credits. New accounts get 100 free credits.

Which file types are supported?

PDF, Word, Excel, PowerPoint (including 97-2003 formats), CSV, text, HTML, RTF, OpenDocument, EPUB and bibliography files on every model, plus PNG, JPG, TIFF, BMP and HEIF images on Auto, Pro and Max.

Use PDF Vector somewhere else

Start Processing Documents Today

Get 100 free credits to try all our APIs. No credit card required.

Free forever tier · No credit card · Cancel anytime