MCP server

Verify PDFs from inside any AI agent

Two surfaces, one product. A remote MCP server at mcp.verifypdf.com/mcp with six tools: two public knowledge-bank tools (search_blog and read_blog_post — long-form fraud detection writing, no key, rate-limited per IP) and four customer tools that drive the paid pipeline (API key required) for Claude, Claude Code, ChatGPT, Perplexity, Gemini and any HTTP-speaking agent. Plus WebMCP on every page of verifypdf.com so an agent browsing the site (Chrome 146 beta + Cloudflare Browser Run) can call six in-page tools directly without leaving the chat.

https://mcp.verifypdf.com/mcp

See the interactive demo

Works with Claude ChatGPT Perplexity Gemini

How it works

Three steps. Most of them happen automatically the first time your agent talks to us.

1

The agent discovers the server

AI assistants read the server card at /.well-known/mcp.json and pick up the MCP endpoint URL automatically.

Inspect the server card
2

A tool call is routed to VerifyPDF

The agent sends JSON-RPC requests over Streamable HTTP to the MCP worker, which delegates to the same API our dashboard uses.
3

Structured JSON comes back

Every response is typed. The agent can quote the verdict, render risk indicators or hand the result off to its next step.

Tool manifest

Six tools, live in production. Two public knowledge-bank tools (search_blog, read_blog_post) work without an API key and are rate-limited per IP. The four customer tools wrap the paid pipeline and require Authorization: Bearer <api_key> on every call. Log into secure.verifypdf.com and open the Developers section to create one.

Tool	Description	Auth
`search_blog`	Search the VerifyPDF knowledge bank — articles on document fraud detection, AI-generated forgeries, payslips, bank statements, tenant screening, KYC/AML, invoice fraud and related verification topics. Returns ranked hits with snippets.	None
`read_blog_post`	Fetch the full body of one knowledge-bank article by slug. Pair with search_blog to find the right slug first.	None
`auth_status`	Confirm an API key is configured for the session. Returns a last-4 fingerprint. Does not call the backend.	Bearer
`submit_document`	Runs the full fraud and tamper analysis on a PDF and returns a structured result with fraud_risk, warnings, presigned URLs and modified-page images. Consumes one document from your plan allowance (or one prepaid document if the monthly allowance is exhausted). Polls every 5s for up to 90s.	Bearer
`get_document`	Fetch the analysis result for a previously submitted document by document_id or custom_id.	Bearer
`delete_document`	Soft-delete a previously submitted document.	Bearer

Configure your VerifyPDF API key

The four customer tools all use the same scheme: send your VerifyPDF API key on every MCP request as an Authorization header. Most MCP clients have a per-server headers field; paste the value below. The two public knowledge-bank tools (search_blog and read_blog_post) work with or without a header — the header is only consulted by the customer tools.

Authorization: Bearer YOUR_VERIFYPDF_API_KEY

Log into secure.verifypdf.com and open the Developers section to generate keys.
Call auth_status first to confirm the key is wired up.
The key is forwarded to the same REST API your existing integrations use. Quota and prepaid balances apply normally.

WebMCP: in-page tools

Every page on verifypdf.com registers six tools through the WebMCP browser API (navigator.modelContext.registerTool). An AI agent navigating the site in a supported browser can call them directly, without scraping HTML or hopping to a remote endpoint.

How it works

Agent navigates to verifypdf.com in a WebMCP-capable browser (Chrome 146 beta inside a Cloudflare Browser Run lab session today; broader support rolling out).
The page detects 'modelContext' in navigator and registers the six tools at idle.
Agent calls navigator.modelContext.listTools() to discover them, then executeTool() to run any one.
Tools run in the page's JavaScript. They post to the same endpoints the human-facing UI uses.

Six in-page tools, all unauthenticated

verify_document: free-tier PDF check (one per business email)
get_pricing: plans, bundles and a recommendation for a monthly volume
list_languages: supported report locales
read_blog_post: fetch a blog post body by slug
book_demo: sales demo request
contact_sales: sales, support or partnership inquiry

The four customer-facing paid-pipeline tools (submit_document, get_document, etc.) are NOT exposed via WebMCP because the page does not hold a VerifyPDF API key. Use the remote MCP server with Authorization: Bearer for those.

Spec: webmachinelearning.github.io/webmcp · Cloudflare reference: developers.cloudflare.com/browser-run/features/webmcp

Connect from your AI assistant

One path for every client: paste the endpoint URL and add an Authorization header with your VerifyPDF API key.

How to connect Claude Code to VerifyPDF MCP

Recommended for developers

Anthropic’s CLI for Claude. One command adds the server to the current project; add `--scope user` to share it across every project on your machine, or `--scope project` to write a `.mcp.json` you can commit to your team repo.

Install with your VerifyPDF API key (all six tools)

One-line install

bash

claude mcp add --transport http verifypdf https://mcp.verifypdf.com/mcp --header "Authorization: Bearer YOUR_VERIFYPDF_API_KEY"

Or commit .mcp.json with an env var (safe for shared repos)

.mcp.json

{
  "mcpServers": {
    "verifypdf": {
      "type": "http",
      "url": "https://mcp.verifypdf.com/mcp",
      "headers": {
        "Authorization": "Bearer ${VERIFYPDF_API_KEY}"
      }
    }
  }
}

Export VERIFYPDF_API_KEY in your shell. Claude Code expands ${VAR} references at load time, so the secret never lives in source control.

Verify: Open Claude Code in the project and run /mcp. The six VerifyPDF tools appear in the list.

How to connect Claude (claude.ai web and desktop) to VerifyPDF MCP

Anthropic’s web and desktop apps. Add a custom connector once, use it across any new conversation.

Server URL

https://mcp.verifypdf.com/mcp

Authorization header

Bearer YOUR_VERIFYPDF_API_KEY

Open Settings, then Connectors.
Choose "Add custom connector".
Paste the server URL above into the connector URL field.
Expand the headers section and add a header: name `Authorization`, value `Bearer YOUR_VERIFYPDF_API_KEY`.
Save. The six tools appear inside any new conversation.

How to connect ChatGPT to VerifyPDF MCP

Available to Teams, Enterprise and Edu workspaces. A workspace admin sets it up once for everyone.

Server URL

https://mcp.verifypdf.com/mcp

Custom header

Authorization: Bearer YOUR_VERIFYPDF_API_KEY

An admin opens Settings, then Custom Connectors.
Click "Add custom connector".
Paste the server URL above.
Add a Custom Header with name `Authorization` and value `Bearer YOUR_VERIFYPDF_API_KEY`.
Save. The six tools appear in any chat across the workspace.

How to connect Perplexity to VerifyPDF MCP

Perplexity supports MCP servers in its Pro and Enterprise plans.

Server URL

https://mcp.verifypdf.com/mcp

Authorization header

Bearer YOUR_VERIFYPDF_API_KEY

Open Settings, then Connectors.
Pick "Custom" and add a new MCP server.
Paste the server URL above.
Add an Authorization header with value `Bearer YOUR_VERIFYPDF_API_KEY`.
Ask any question that involves verifying a PDF and Perplexity will call the tool.

How to connect Gemini CLI to VerifyPDF MCP

Google’s open-source CLI for Gemini. Installs over npm and supports MCP over streamable HTTP natively.

Install with your VerifyPDF API key (all six tools)

Install the CLI

bash

npm install -g @google/gemini-cli

Add ~/.gemini/settings.json with a headers block

~/.gemini/settings.json

{
  "mcpServers": {
    "verifypdf": {
      "httpUrl": "https://mcp.verifypdf.com/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_VERIFYPDF_API_KEY"
      }
    }
  }
}

Gemini AI Studio and Vertex AI Agent Builder MCP support is rolling out separately. The same endpoint URL applies once it ships.

Verify: Run `gemini`, then type /mcp to confirm the tools are loaded.

Try it from the command line

Same wire format your AI assistant uses. Useful for smoke testing or wiring up your own agent.

bash

curl -X POST https://mcp.verifypdf.com/mcp \
  -H 'content-type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Trust and safety

The same guarantees we make on the REST API, applied to agent traffic.

Rate limited per IP so a noisy agent cannot block a polite one.
submit_document caps at 10 MB base64. Larger files belong on the dashboard or REST API.
No document retention beyond the verification window. Once the report is returned, the file is gone.
Every tool call is logged for audit, the same way REST API requests are.
Every tool forwards your API key as the API-KEY header to the same REST endpoints your existing integrations use. We never store the key. It lives only for the duration of the request.

FAQ

What is the VerifyPDF MCP server?

A remote Model Context Protocol server at https://mcp.verifypdf.com/mcp that exposes six VerifyPDF tools to AI agents over Streamable HTTP. Two are public (search_blog and read_blog_post — the VerifyPDF knowledge bank on document fraud detection) and work without a key, rate-limited per IP. Four are customer tools (auth_status, submit_document, get_document, delete_document) that wrap the paid REST API and require a VerifyPDF API key sent as Authorization: Bearer on every request.

How do I connect Claude Code to VerifyPDF MCP?

Run `claude mcp add --transport http verifypdf https://mcp.verifypdf.com/mcp --header "Authorization: Bearer YOUR_VERIFYPDF_API_KEY"` in your project. Verify with /mcp inside Claude Code.

What is MCP?

MCP is the Model Context Protocol, an open standard that lets AI assistants talk to outside tools and data sources. Claude, Claude Code, ChatGPT, Perplexity and Gemini all support it. You point the agent at the endpoint URL and it learns what tools the server exposes and how to call them.

How is this different from your REST API?

Same data, different audience. The REST API is for engineers building integrations. The MCP server is for AI agents calling tools inside a chat. Under the hood, both reach the same verification engine. The MCP tools (submit_document, get_document, delete_document) are thin wrappers around the REST API and forward your API key as the API-KEY header.

How do I get a VerifyPDF API key?

Log into your account at secure.verifypdf.com and open the Developers section. Create a key, copy it, and paste it as the Bearer value in your MCP client. The key never touches the MCP worker beyond a single request: it is forwarded to the REST API and discarded.

Will my documents be retained?

No. We process the PDF, return the verdict and discard the file. The behaviour is identical to the public REST endpoint.

What about authentication?

Two tiers. The four customer tools (auth_status, submit_document, get_document, delete_document) all require an Authorization: Bearer header carrying your VerifyPDF API key — generated in the Developers section of secure.verifypdf.com. The worker forwards it to the same REST API your existing integrations use; quota, prepaid balance and audit logging apply normally. The two public knowledge-bank tools (search_blog, read_blog_post) work with or without a header and are rate-limited per IP (60 calls per minute).

What is the VerifyPDF knowledge bank?

Long-form articles by the VerifyPDF team covering document fraud detection, AI-generated forgeries, payslip and bank-statement verification, tenant screening, KYC and AML compliance, invoice fraud, and related topics. Agents reach it via two unauthenticated MCP tools: search_blog (ranked keyword search across the whole knowledge bank) and read_blog_post (full article body for a given slug). Any time a user asks how document fraud works or how to verify a specific document type, the answer is usually one search_blog call away — no API key needed.

Can I build my own agent on top of these tools?

Yes. The endpoint speaks the open Model Context Protocol over Streamable HTTP, so anything that implements MCP can connect. Point your client at https://mcp.verifypdf.com/mcp (add an Authorization: Bearer header for the customer tools), call tools/list to discover the six tools and their JSON schemas, then call tools/call to invoke them. The MCP specification is published at modelcontextprotocol.io.

Want a guided tour first?

Talk to a human, watch a demo or send us the integration you have in mind. We will help your agent get set up the right way.

Watch the product demo Talk to sales

Verify PDFs from inside any AI agent

How it works

The agent discovers the server

A tool call is routed to VerifyPDF

Structured JSON comes back

Tool manifest

Configure your VerifyPDF API key

WebMCP: in-page tools

How it works

Six in-page tools, all unauthenticated

Connect from your AI assistant

Try it from the command line

Trust and safety

FAQ

Want a guided tour first?