--- title: "/mcp: Native AI access for your domain | Serpwise" description: "Serve a Model Context Protocol endpoint on your domain so AI agents can query content through structured tools, with bot policies and agent analytics." url: "https://serpwise.ai/mcp-server/" source: "https://serpwise.ai/mcp-server/" --- MCP server · AI access control # yourdomain.com /mcp Serpwise runs a native Model Context Protocol server on every domain it fronts. AI agents including ChatGPT, Claude, and Perplexity can search your pages, ask grounded questions, and read clean markdown through 7 structured tools. No API key. No plugins. Zero origin changes. [Book a demo](/demo/) [Start free trial](https://cloud.serpwise.ai/auth/sign-up) any-mcp-client - connected › connect https://yourdomain.com/mcp ✓ serpwise-site-content · 7 tools available › ask "Do you ship to Norway?" ✓ grounded answer · 3 citations › search\_content "return policy" ✓ 5 ranked pages · clean markdown › Live within 60 seconds Global edge delivery after approval Zero origin code changes Your CMS and release flow stay untouched Preview and rollback Every change is controlled and reversible 01 · Why it matters ## AI agents already read your site. Badly. Scraping loses structure Agents parsing raw HTML miss context, mangle tables, and hallucinate details. Their users may then hear those details as facts about your business. Answers replace clicks ChatGPT, Claude, and Perplexity answer questions directly. If they can't read you cleanly, they cite whoever they can read instead. MCP is the open standard The Model Context Protocol is how AI agents connect to data sources. Serpwise makes your site a first-class source and serves it from the edge. 02 · The tools ## Seven structured tools. One toggle. Every tool returns typed, structured output an agent can reason over instead of a wall of HTML it has to guess at. `search_content` Content search Semantic search across every indexed page, including blog posts, docs, policies, and FAQs. Results are ranked by relevance. `ask` Grounded Q&A Ask a natural-language question and get a synthesized answer with citations from your indexed content, not the open web. `get_page` Page retrieval The full content of any page as clean markdown with metadata, ready for AI reasoning, stripped of layout noise. `list_pages` Page index Enumerate every indexed page, paginated, so agents can map your whole site before drilling into specifics. `search_products` Product search Natural-language catalog search for sites with structured product data. Results include title, price, availability, brand, and category. `get_product_details` Product details A full enriched product profile with specs, highlights, and attributes returned as one structured object. `list_categories` Category browser The complete product category hierarchy, so agents understand how a catalog is organized before they search it. The three catalog tools activate automatically when structured product data is available on your site. Content tools work across SaaS docs, portfolios, publishers, stores, and other sites. Under the hood: built for relevance and speed 1 Retrieve Hybrid search Every query runs dense and sparse semantic search in parallel, then fuses the results. Meaning is matched without losing exact terms. 2 Rerank Custom reranker A dedicated reranking model, served on our own GPU infrastructure, re-scores every candidate so the most relevant passages rise to the top. 3 Serve Edge-cached Popular queries are cached at the edge and answered in milliseconds. Agents get answers at conversation speed. 03 · Open by default ## No key. No signup. No friction. Consumer AI agents cannot hold per-site API keys, so we do not ask for one. Any MCP client connects with just the URL and reads what is already public on your site. Nothing more. Guarded at the edge - Per-IP rate limiting before any tool runs - Scoped to the one domain it serves and nothing else - Read-only by design: no write tools exist - Serves only content your public site already shows - One-click switch to keyed-only access at any time Auto-discoverable `/mcp` The MCP endpoint itself, live on your own domain. `/.well-known/mcp.json` The standard manifest MCP clients look for. It lists the server, endpoint, and tools. `/llms.txt` A plain-text pointer following the emerging llms.txt convention, so LLM crawlers find the structured door. Every call is logged to your Agent Traffic dashboard, whether it is anonymous or keyed. Open access never means blind access. 04 · AI access control ## You decide which AI reads you, and how. The MCP server is the front door. AI Access Control governs the rest of the house: per-crawler policy, the format agents are served, and the signals you publish about search, AI input, and training. Per-bot policy Allow, monitor, or block crawlers by name, including GPTBot, ClaudeBot, PerplexityBot, CCBot, and Google-Extended. Training crawlers can be denied outright. Search and user-triggered agents stay observable. Markdown for agents Serve approved AI agents clean, token-efficient markdown instead of your HTML and JavaScript bundle. It is the same content without the noise. Content signals Publish machine-readable signals for search, AI input, and AI training, so compliant crawlers know your terms before they fetch a single page. Agent traffic analytics See which agents call which tools, what they ask, and which queries come back empty. These content gaps come straight from real AI demand. Bot policy Googlebot Search Allow OAI-SearchBot AI search Allow ChatGPT-User User-triggered Allow PerplexityBot AI search Monitor ClaudeBot Training Monitor GPTBot Training Block CCBot Training Block 05 · How it works ## Live in minutes. Zero code changes. 01 Flip the toggle Enable the MCP server in your domain's AI Access settings. The edge starts serving /mcp, the manifest, and llms.txt immediately. 02 We index and serve Your crawled pages are embedded for semantic search. The gateway answers every agent call at the edge, so your origin never sees MCP traffic. 03 Agents find you MCP clients auto-discover the endpoint and start querying. You watch every call, question, and content gap in Agent Traffic. FAQ ## Questions, answered. 01 What is MCP, and why should my site have it? + - The Model Context Protocol is an open standard that lets AI agents connect to data sources through typed, structured tools. Without it, agents scrape your HTML and lose structure along the way. With it, they call tools like search\_content or ask and get clean, ranked, citable results. Your site becomes a first-class data source for every AI assistant. 02 Which AI agents can connect? + - Any MCP-compatible client: Claude, ChatGPT, Perplexity-class agents, IDE assistants like Cursor, and custom agents built on the MCP SDK. The discovery endpoints (/.well-known/mcp.json and /llms.txt) help them find your server automatically. 03 Do agents need an API key? + - No. Read access is open by default because consumer AI agents cannot hold per-site secrets. Calls are rate-limited per IP at the edge, scoped to your domain only, and expose nothing beyond content that is already public. You can switch to keyed-only access at any time. 04 Do I need to change my website? + - No. The endpoint is served entirely by the Serpwise edge in front of your site. Your origin never receives MCP traffic, and no plugin, theme edit, or code change is required. 05 What exactly can agents read? + - Nothing more than what your public site already shows. Agents can read crawled pages as clean markdown, search them semantically, and get grounded answers with citations. Sites with structured product data also expose catalog search. Everything is scoped to your domain, never anyone else's. 06 Can I block AI crawlers instead? + - Yes. The same AI Access panel lets you allow, monitor, or block individual crawlers and publish content signals covering search, AI input, and AI training. Welcome the agents that send you customers; deny the ones that don't. Works with the rest of Serpwise [GEO Optimization → \ Make your pages the answer AI assistants choose to cite.](/product/geo-optimization/) [AI Visibility → \ Track how ChatGPT, Claude, and Perplexity talk about you.](/product/ai-visibility/) [Analytics MCP → \ The other MCP: query your own Serpwise data from your AI tools.](/product/analytics-mcp/) [Security & trust → \ How the edge stays fail-open, data-minimized, and out of the way.](/security/) Get started ## Put your site on /mcp. One toggle in the dashboard puts it live on any stack in minutes. Every AI agent can finally read you properly. [Book a demo](/demo/) [Start free trial](https://cloud.serpwise.ai/auth/sign-up)