Documentation

What is WhipNode?

WhipNode is a context relay for AI agents. Upload a screenshot, file, or document at whipnode.com, and your AI tool can read it instantly - with OCR text extraction, structured outputs, and your notes attached. You can also give your agent a URL and WhipNode will screenshot it for you.

Getting Started

Step 1 - Create an account

Sign in with your email. No password needed - we send you a magic link.

Step 2 - Create an API key

Go to the API Keys tab in your dashboard and create a key. Save it - it's only shown once.

Step 3 - Install

Run this in any terminal window:

claude mcp add whipnode --transport http https://whipnode.com/api/v1/mcp --header "X-API-Key: YOUR_API_KEY" -s user

The -s user flag installs WhipNode globally - available in every project, not just the current directory. Then add WhipNode instructions to your global ~/.claude/CLAUDE.md so Claude knows when to use each tool. See the API Keys tab for the full instructions block to copy.

Your API key is used once to register the device. After that, WhipNode assigns a secure session token automatically - the key is never sent again. For Codex CLI and other tools, see the setup instructions in your dashboard.

Step 4 - Upload and share

Upload a screenshot or file at whipnode.com/new. Paste the URL into your Claude Code session. Your agent reads it automatically using the whipnode_fetch tool.

How It Works

1. You upload

Paste a screenshot, drag a file, or upload via the API. Add an optional note describing what you need.

2. WhipNode processes

Images are scanned with OCR to extract text. Every upload is converted into five formats - JSON, Markdown, YAML, plain text, and a prompt pack optimized for AI agents. Files are hashed with SHA-256 for integrity.

3. Your agent reads

If you connected Claude Code via MCP, it can read WhipNode content natively using the whipnode_fetch tool. For other tools, paste the URL - the agent fetches it directly.

MCP Tools

When connected via MCP, your AI tool gets six tools automatically:

ToolWhat it does
whipnode_fetchRead any WhipNode item - OCR text, files, notes
whipnode_uploadUpload a file and get a shareable URL
whipnode_captureScreenshot a live URL at any viewport (desktop, mobile, tablet, etc.)
whipnode_renderRender HTML at multiple viewports - responsive design comparison
whipnode_listList your recent uploads
whipnode_helpShow available tools and common workflows

URL Capture

Give WhipNode a URL and it takes a screenshot using a headless browser on a global edge network. The screenshot is processed with OCR and converted to all five output formats - just like a file upload. A sitemap of the page's navigation links is also extracted automatically.

Capture multiple viewports in one request (e.g. desktop + tablet + mobile) for responsive design checks. Google Sheets URLs are automatically exported as CSV instead of screenshotted - giving your agent the actual data.

Viewport presets: desktop (1280x800), laptop (1024x768), tablet (768x1024), mobile (375x812), 1080p, and 4K. Full-page capture optional.

Use cases: sharing staging sites with your agent, checking what a live page looks like, capturing error pages for debugging, reviewing responsive layouts, importing spreadsheet data.

Multi-Agent Workflows

WhipNode works as a shared context layer between AI agents. Any agent with WhipNode tools can upload files and read other agents' uploads - regardless of which machine they're on.

How agents find each other

Agents on the same WhipNode account share a common upload history. Any agent can call whipnode_list to see recent uploads from all agents on the account - across all machines and sessions. The account (API key) is the shared namespace.

Same machine

Two Claude Code sessions on the same machine, both connected to WhipNode with the same API key, can see each other's uploads immediately. Agent A uploads a file, Agent B calls whipnode_list and fetches it. Or paste a WhipNode URL from one session into the other.

Different machines

Agents on different machines work exactly the same way - WhipNode is cloud-hosted. Install WhipNode on Machine B with the same API key (or a different key on the same account), and both agents share the same upload history. An agent on your laptop and an agent on a CI server can exchange context through WhipNode URLs.

Single-session orchestration

The simplest multi-step workflow runs in one session: capture a site, analyze the results, upload a report. No URL passing needed - the agent chains its own tool calls. Example: capture a staging deploy, extract the OCR text, compare against requirements, upload findings.

Current limitations

Today, agents discover each other's uploads by polling whipnode_list or by having URLs passed to them by a human. There is no push notification or real-time channel yet - if Agent B needs to react to Agent A's upload, it needs to check periodically or be told the URL.

Responsive Rendering

Upload HTML or provide a preview item, and WhipNode renders it at multiple viewports simultaneously. See how your page looks on desktop, tablet, and mobile side by side.

The share page shows a gallery view with viewport tabs, device icons, resolution labels, and download buttons per viewport.

Security

Device-based authentication

When you connect Claude Code (or any MCP client), your API key is used once to register the device. WhipNode assigns a secure session token (Mcp-Session-Id) that the client stores and uses for all future requests. The API key is never sent again.

You can see all connected devices in the Devices tab of your dashboard. Revoke any device instantly - the next request from that device will fail and need to re-authenticate.

Key rotation

Rotate an API key from the dashboard. A new key is generated, and your old key continues working for 15 minutes (grace period) while you update your agents. No downtime.

Data handling

  • All uploads expire after 24 hours and are permanently deleted
  • Files are hashed with SHA-256 for integrity verification
  • API keys are stored as SHA-256 hashes - raw keys never stored
  • Session tokens hashed at rest, 90-day expiry
  • Output renderers include safety preambles warning agents to treat user content as untrusted
  • Prompt injection scanner detects 13 common attack patterns

Output Formats

Every upload generates five formats, each at its own URL:

FormatURL PatternBest for
JSON/api/v1/items/:idProgrammatic access, full metadata
Markdown/api/v1/items/:id.mdHuman reading, LLM-friendly
YAML/api/v1/items/:id.yamlAutomation, config-style output
Plain Text/api/v1/items/:id.txtUniversal fallback
Prompt Pack/api/v1/items/:id/prompt.mdAI agents - includes context, task, and file data

Further Reading