AI Agent Protocol
Wrenda transforms your website into an AI-native resource. Discoverable by AI agents via MCP protocol, readable by LLMs as clean markdown -- automatically.
Compatible with
GPT-4oClaude 3.5Gemini 1.5LangChainAutoGPTCrewAITwo ways AI agents interact with your site
Wrenda automatically detects what kind of AI is asking and delivers the right format.
Feature 01
MCP Server -- Tool Discovery & Execution
Wrenda exposes a standards-compliant Model Context Protocol server on every customer domain. AI agents built with any framework can discover your site's capabilities and call tools in real time -- querying products, docs, pricing, and inventory without scraping.
- Tool manifest served at /.well-known/mcp.json
- JSON-RPC 2.0 endpoint at /.well-known/mcp
- Custom tools: search, filter, lookup, compare
- Rate limiting and auth token support
- Works with LangChain, AutoGPT, CrewAI, and more
yoursite.com/.well-known/mcp.json
{
"protocol": "mcp/1.0",
"name": "Your Site Tools",
"tools": [
{
"name": "search_products",
"description": "Search product catalog",
"inputSchema": {
"type": "object",
"properties": {
"query": { "type": "string" },
"category": { "type": "string" },
"maxResults": { "type": "number" }
}
}
},
{
"name": "get_pricing",
"description": "Get current pricing for SKU"
}
]
}How an AI agent interacts
AI Agent
GPT / Claude / Gemini
Discovers tools
GET /.well-known/mcp.json
Calls search_products
POST /.well-known/mcp
Structured JSON
{ "products": [...] }
Tool response
{
"result": {
"products": [
{ "id": "A1", "name": "Apex Pro",
"price": 189.99, "stock": 42 },
{ "id": "A2", "name": "Velocity X",
"price": 149.99, "stock": 8 }
],
"total": 47
}
}Feature 02
Markdown Rendering -- Clean LLM Context
HTML (what LLMs struggle with)
<div class="container">
<div class="nav-wrapper">
<script src="/bundle.js"></script>
<div data-react-root></div>
</div>
<!-- 47 divs of nothing -->
<div class="content">
<h1 class="heading-xl">
Marathon Shoes
</h1>
</div>
</div>Markdown (clean LLM context)
# Marathon Shoes Premium running footwear designed for long-distance performance. ## Key Features - Carbon fiber plate - 40mm stack height - 6.8oz race weight ## Specifications | Attribute | Value | |-----------|-------| | Drop | 8mm | | Width | B–2E |
When an AI system sends Accept: text/markdown in its request, Wrenda automatically strips all HTML markup and delivers a clean, structured markdown document. No JavaScript, no nav boilerplate, no cookie banners -- just the content LLMs need.
Use cases
- RAG Pipelines -- Feed clean content into vector stores without HTML noise
- LLM Summarization -- Token-efficient markdown means lower costs and better summaries
- Agent Memory Systems -- Store meaningful content, not markup soup
- Citation & Grounding -- LLMs accurately cite structured markdown over HTML
Trigger it from any LLM client
curl https://yoursite.com/products/shoes \ -H "Accept: text/markdown" \ -H "User-Agent: MyRAGBot/1.0" # Returns clean markdown, no HTML
Zero configuration required
Point your domain at Wrenda. Everything else is automatic.
Step 01
Wrenda detects the request
Incoming requests are fingerprinted in under 1ms. AI agent user agents, Accept headers, and JSON-RPC calls are identified precisely -- real users are never affected.
Step 02
Selects the right format
MCP JSON-RPC for agentic AI clients. Clean markdown for LLMs ingesting content. Enriched semantic HTML for crawlers like GPTBot and ClaudeBot. The right format, automatically.
Step 03
Serves optimized content
Responses are edge-cached at 300+ Cloudflare locations. Cold transformations run in under 50ms. Subsequent requests serve from cache in single-digit milliseconds.
Step 04
Agents understand your site
AI tools can now query your inventory, cite your articles, recommend your products, and pull your documentation -- integrating your site into the agentic web.
Built for the agentic web
AI agents are already browsing, buying, and building. Make sure they find you.
Shopping AI Agents
AI agents researching products call your search_products and get_pricing tools directly. No scraping, no hallucination -- live data from your catalog.
- Get recommended in ChatGPT Shopping queries
- Real-time inventory and pricing via MCP tools
- Structured product data agents actually trust
RAG & Knowledge Bases
LLMs pulling content for retrieval-augmented generation receive clean markdown. No HTML parsing, no boilerplate stripping, no content extraction hacks.
- Markdown served on Accept: text/markdown
- Token-efficient -- reduces embedding costs
- Consistent structure across all pages
AI Documentation Bots
Support bots and AI assistants can call your docs tools to get accurate, up-to-date answers directly from your site -- not from stale training data.
- search_docs and get_article MCP tools
- Always-current answers from live content
- Reduces hallucination in support workflows
It just works
Standard MCP protocol. Any agent can discover and use your tools in seconds.
# Step 1 — AI agent discovers your tools GET yoursite.com/.well-known/mcp.json # Step 2 — Agent calls your search tool POST yoursite.com/.well-known/mcp Content-Type: application/json { "jsonrpc": "2.0", "method": "tools/call", "params": { "name": "search_products", "arguments": { "query": "marathon shoes" } }, "id": 1 } # Step 3 — Your site responds with live structured data { "jsonrpc": "2.0", "result": { "products": [ { "id": "A1", "name": "Apex Pro", "price": 189.99, "stock": 42 }, { "id": "A2", "name": "Velocity X", "price": 149.99, "stock": 8 }, { "id": "A3", "name": "TrailBlazer Z", "price": 219.99, "stock": 15 } ], "total": 47, "query": "marathon shoes" }, "id": 1 }
<50ms
Tool response time
MCP 1.0
Protocol version
JSON-RPC 2.0
Transport standard
Get Started
Ready for the agentic web?
Enable WebMCP in minutes. No code changes, no infrastructure -- just point your domain and go.
Enable WebMCP14-day free trial -- No credit card required