Be found by AI.
The open standard for the agentic web.

You need Python 3.10+, a domain, and a VPS. The entire server is one file.

pip install fastmcp

{
  "name": "Your Business",
  "description": "What you do in one sentence.",
  "services": [
    { "name": "Service Name", "price_range": "$X–$Y" }
  ],
  "location": { "city": "Boise", "state": "ID" }
}

A PUBLICMCP listing gives your business a structured profile that AI assistants — ChatGPT, Claude, Google's AI — read directly. When someone asks an AI for a business like yours, it answers with your real, current facts instead of guessing from scraped web pages. Stand it up yourself for free, or have us do it for you.

Three tool calls to library.publicmcp.org — from graduate-level device physics to OSHA regulatory law to EUV research spanning four universities. Every response cites the primary source. No scraping. No API key.

{ "count": 2, "results": [
  { "item_id": "micron_20260605_194945",
    "title": "MOS Transistor",
    "description": "Full textbook chapter — I-V characteristics, threshold voltage, body effect.",
    "item_type": "book", "institution": "Micron Technology, Inc." }
]}

{ "title": "MOS Transistor", "institution": "Micron Technology, Inc.",
  "item_type": "book", "source": "micron-training",
  "full_text": "...Idsat = W/(2mL) × Coxe × µns × (Vgs − Vt)²  [Eq. 6.6.6]
               m = 1 + 3Toxe/Wdmax ≈ 1.2 (bulk-charge factor)
               [71,015 characters — complete primary source]..." }

{ "title": "OSHA 29 CFR 1910.147 — The Control of Hazardous Energy (Lockout/Tagout)",
  "institution": "U.S. Department of Labor, OSHA", "item_type": "standard",
  "full_text": "...## 1910.147(d) — Application of Control\n
               1. Preparation for Shutdown...\n  2. Machine or Equipment Shutdown...\n
               3. Machine or Equipment Isolation...\n  4. Lockout or Tagout Device Application...\n
               5. Stored Energy...\n  6. Verification of Isolation..." }

{ "count": 20, "results": [
  { "title": "Extreme ultraviolet lithography reaches 5 nm resolution",
    "institution": "Paul Scherrer Institute", "source": "micron-training" },
  { "title": "Maskless extreme ultraviolet lithography",
    "institution": "Stanford University", "source": "openalex" },
  { "title": "Time exposure performance of Mo-Au alloys for EUV collector optics",
    "institution": "University of Illinois Urbana-Champaign", "source": "openalex" },
  { "title": "Material-specific high-resolution table-top EUV microscopy",
    "institution": "Friedrich-Schiller-Universität Jena", "source": "micron-training" },
  // ... 16 more
]}

Students using AI to research shouldn't get hallucinated answers because a library's content isn't structured for machine consumption. PUBLICMCP EDU gives public libraries and educational nonprofits a permanently free hosted server — no DNS changes, no technical setup, no cost. Ever.

Verified educational nonprofits and public libraries only. One email is all it takes.

The PUBLICMCP Public Library is an open content layer built on the same infrastructure as the standard. Any AI agent can search and retrieve primary sources across six collections: peer-reviewed CS and engineering papers from 50 top universities via OpenAlex (165,876 items), digitized historical records from the New York Public Library (18,541 items), 187 curated semiconductor manufacturing training resources for new and aspiring engineers, and scientific preprints from arXiv — all served via MCP tools, no API key required.

No paid plan. No API key. Works with any free Claude account — once connected, your Claude has access to 184,000+ primary sources in every conversation.

https://library.publicmcp.org/mcp

https://mcp.publicmcp.org/mcp

I'm studying semiconductor manufacturing. Show me the training library sections and find beginner content on photolithography.

PUBLICMCP is an open standard for public-facing business and nonprofit identity MCP servers. It defines a required set of tools — get_info, get_services, get_location — that any organization can expose so AI agents can discover and query them without scraping or authentication.

Yes. The standard is open — self-host a compliant server for the cost of a VPS (about $5–$20/month), or have Salt Creative host a basic listing for you free. An optional Verified tier — a one-time human review of your license and credentials — is a paid add-on.

schema.org embeds structured data in HTML for search engine crawlers. PUBLICMCP exposes live, queryable tools that AI agents call directly via the Model Context Protocol — no scraping required. Responses can include real-time data like current pricing and availability, and update instantly without touching your website.

Any agent that supports the Model Context Protocol: Claude, GPT, Gemini, Ollama, and custom agents. If the agent can connect to an MCP server, it can query any PUBLICMCP-compliant server natively via /mcp.

For browser chat users: Claude.ai is the only mainstream browser UI with native MCP connector support. Developers using the Gemini API, Ollama, or any MCP-compatible framework connect via /mcp. ChatGPT integrations can use the REST endpoint at /tools/{tool_name}.

Open Claude.ai → click your profile icon → Settings → Connectors → Add MCP Server → paste https://library.publicmcp.org/mcp → Save. That's it. No paid plan required — a free Claude account works. Your Claude will have access to 184,000+ primary sources — including the semiconductor manufacturing training library — in every conversation from that point forward.

For self-hosting, you need Python 3.10+, a VPS, and a domain. FastMCP is the reference runtime — install it with pip install fastmcp, configure your business.json, and deploy. If that's out of reach, Salt Creative offers fully managed deployments typically live within 48 hours.

Required tools (get_info, get_services, get_location) must be implemented for spec compliance. Recommended tools (get_portfolio, get_markets, get_products, search_site_content, get_page_content, list_site_pages) are strongly encouraged and included in the reference implementation but are not mandatory for a compliant server.

Every registry entry requires a live compliant server at a public URL. If you can't self-host, Salt Creative's managed hosting service covers setup, hosting, and maintenance so your business appears in the registry without running any infrastructure yourself.

PUBLICMCP servers are intentionally public and read-only — they expose only what you choose to include in your business.json. No authentication is required for agents to query your server, and no customer data or private information should ever be included. Think of it as a public business profile, not a database.

Yes. PUBLICMCP is an open standard — anyone can implement a compliant server on any stack. The standard is designed to evolve alongside the Model Context Protocol, which is already incorporated into official Anthropic, OpenAI, and Google AI SDKs. FastMCP, the leading MCP framework with 1M+ daily downloads, serves as the reference runtime. The foundation is solid.