sebbi Proven MCPThe MCP registry that actually checks

← The registry

The Proven MCP Standard PM-1

A server is Proven when all five hold. Each one is something you can check yourself.

1. It answers

An HTTPS endpoint that completes MCP initialize and tools/list (protocol 2025-06-18, plain JSON or event stream), within 12 seconds, with a valid certificate, no redirects, on a public address.

2. It is clean

No high findings in anything the AI reads: tool names, titles, descriptions, every string in the schemas, and the server's instructions. High findings are invisible characters; instructions to ignore rules, hide things from the user, act first, read secret files, send data elsewhere, reveal the conversation, change identity or take over other tools; and anything that looks like a live key. Medium and low findings lower the score but don't block it. Every finding is published with the exact words.

3. It is pinned

The PM-1 fingerprint of the tool list is sealed into sebbi.pro's chain, which is anchored in Bitcoin. Each tool also has its own fingerprint, so a change shows exactly which tool moved.

import hashlib, json

KEEP = ("title", "description", "inputSchema", "outputSchema", "annotations")

def pm1_fingerprint(tools):            # tools = the "tools" list from tools/list
    clean = []
    for t in tools:
        c = {"name": str(t.get("name") or "")}
        for k in KEEP:
            if k in t and t[k] not in (None, "", {}):
                c[k] = t[k]
        clean.append(c)
    clean.sort(key=lambda c: c["name"])
    text = json.dumps(clean, sort_keys=True, separators=(",", ":"), ensure_ascii=False)
    return hashlib.sha256(text.encode("utf-8")).hexdigest()

# Then ask sebbi:  POST https://sebbi.pro/x/mcpproof/verify
#   {"url": "<server address>", "fingerprint": pm1_fingerprint(tools)}
# MATCH = these are the tools that passed.  MISMATCH = don't trust them.

4. It is watched

Re-tested every 24 hours. A different fingerprint turns the listing CHANGED at once. It is Proven again only when the owner approves the change with their owner key and the new tools pass. Three failed checks in a row mark it DOWN. Every event is sealed and stays on the public record.

5. Anyone can check

Any client computes the fingerprint above from what it actually receives and asks https://sebbi.pro/x/mcpproof/verify. The full registry is at https://sebbi.pro/proven/registry.json.

sebbi Proven MCP · Standard PM-1 · Read the standard · Machine feed · sebbi.pro · Listing is free.

AI news
TechCrunchTrajectory raises $45M to make AI model training accessible to nontechnical experts1hTechCrunchAnthropic warns that it’s very close to RSI capability, aka self-building AI2hThe VergeAnthropic clamps down on Claude’s AI video game clones6hArs TechnicaClaude’s new rules against game cloning won’t stop the AI copying boom6hThe VergeOpenAI pitches data center power plants as a way to protect nearby areas7hThe VergeThe New York Times sues Brave web browser for AI scraping and bypassing paywalls7hMIT Technology ReviewRaising the quality of insights in agentic analytics8hTechCrunchElon Musk throttles Grok Bot limits due to ‘unsustainably high’ costs8hTechCrunchOmen AI adds liquid know-how in $16 million Aquatic Labs acquisition10hTechCrunchOpenAI is finally merging ChatGPT’s Chat and Work tabs11hWiredWhen You Should—and Shouldn’t—Use AI to Search the Web13hWiredHundreds of Mundane AI Apps Are Hiding Sexually Abusive Deepfake Features15hMIT Technology ReviewThe agentic web is here15hWiredChatbots Are Leading Hikers Astray, With Dangerous Consequences1dBBC NewsThese firms you've never heard of are feeding fears the AI bubble could burst1dThe VergeI built an AI Tamagotchi before Meta could2dWiredSpooky New Infostealer Malware Uses AI to ‘Autodox’ Victims2dWiredAn AI Agent Injected Drugs Into a Live Mouse’s Brain2dTechCrunchTrajectory raises $45M to make AI model training accessible to nontechnical experts1hTechCrunchAnthropic warns that it’s very close to RSI capability, aka self-building AI2hThe VergeAnthropic clamps down on Claude’s AI video game clones6hArs TechnicaClaude’s new rules against game cloning won’t stop the AI copying boom6hThe VergeOpenAI pitches data center power plants as a way to protect nearby areas7hThe VergeThe New York Times sues Brave web browser for AI scraping and bypassing paywalls7hMIT Technology ReviewRaising the quality of insights in agentic analytics8hTechCrunchElon Musk throttles Grok Bot limits due to ‘unsustainably high’ costs8hTechCrunchOmen AI adds liquid know-how in $16 million Aquatic Labs acquisition10hTechCrunchOpenAI is finally merging ChatGPT’s Chat and Work tabs11hWiredWhen You Should—and Shouldn’t—Use AI to Search the Web13hWiredHundreds of Mundane AI Apps Are Hiding Sexually Abusive Deepfake Features15hMIT Technology ReviewThe agentic web is here15hWiredChatbots Are Leading Hikers Astray, With Dangerous Consequences1dBBC NewsThese firms you've never heard of are feeding fears the AI bubble could burst1dThe VergeI built an AI Tamagotchi before Meta could2dWiredSpooky New Infostealer Malware Uses AI to ‘Autodox’ Victims2dWiredAn AI Agent Injected Drugs Into a Live Mouse’s Brain2d