For AI agents & assistants

Agent Access

Onchain Diary exposes its full knowledge base — 159 Web3 security articles and 232 glossary terms, in English and Chinese — to AI agents six different ways, all anonymous, read-only and free. Pick a tab below; without JavaScript every route is laid out in full.

Verify it works

Two prompts, two paths. The first checks an MCP connection end to end; the second works in any AI with web access and needs zero setup. If the answer comes back with a real link, the chain works.

MCP-connected assistant

Using the connected Onchain Diary MCP server, call the "search" tool with query "wallet drainer", then call "read_article" on the first result. Summarize its BLUF in 3 sentences and include the article URL. If you cannot reach the server, say so — do not guess.

Any AI, no setup

Read https://theonchaindiary.com/llms.txt and tell me in two sentences what this knowledge base covers. Then pick one article about address poisoning and give me its URL and BLUF summary. If you cannot fetch the URL, say so — do not guess.

Ways to connect

Ask any AI

Readers — zero setup, no MCP needed.

Every page footer carries one-click buttons for ChatGPT, Claude, Perplexity, Gemini and Grok. Each button opens the assistant with a ready-made prompt pointing at this site's llms.txt — you ask, it verifies against the source. The fastest way to put this library to work.

llms.txt

Lightweight agents & crawlers that fetch URLs.

The compact index lists every section and the MCP endpoints; the full version carries all 159 article summaries with dates and tags. Plain Markdown, anonymous GET:

https://theonchaindiary.com/llms.txt
https://theonchaindiary.com/llms-full.txt

MCP server

Developers wiring an assistant, IDE or chat client.

https://theonchaindiary.com/api/mcp

Streamable HTTP · no authentication · read-only · stateless. Discovery file at /.well-known/mcp.json.

Tools

search Full-text search across articles and glossary. Keywords work in English and Chinese — e.g. wallet drainer or 地址投毒. Returns the top 8 matches with title, URL and short description.
read_article Complete markdown of one article, with BLUF summary, tags and FAQ when present.
read_glossary One glossary term: full English definition plus the Chinese translation (译名 + 中文定义).
list_content Table of contents — articles grouped by language and date, or glossary grouped by category.

Boundaries

Start with search, then read full text — it returns the top 8 ranked matches, not an exhaustive list.
Slugs for read_article / read_glossary should come from search or list_content results. Guessed slugs return a clear error, never a fuzzy match.
The data is a static snapshot built at deploy time — brand-new articles appear after the next deployment, not in real time.
Everything is read-only: no authentication, no sessions, nothing can be created or modified.

Client config

Most clients only need the endpoint URL. For clients that take a JSON config:

{
  "mcpServers": {
    "onchain-diary": {
      "url": "https://theonchaindiary.com/api/mcp"
    }
  }
}

Works with Claude Desktop & Claude Code, Cursor, ChatGPT connectors, Windsurf, or any client implementing the MCP Streamable HTTP transport.

Structured JSON

Scripts & agents that prefer raw data over markdown.

The same data the MCP server serves, as plain static JSON — the catalog of every article and glossary term with URLs:

https://theonchaindiary.com/mcp/manifest.json

Each catalog entry then resolves to a per-item document: /mcp/a/<slug>.json for articles and /mcp/glossary/<slug>.json for terms, both carrying the full text. No auth, no rate limit games — it's a static file host.

RSS

News readers & feed-watching agents.

Plain crawl

Any bot — clean HTML, no JS required.

Every page renders server-side as readable HTML with canonical URLs, JSON-LD and hreflang. Start from sitemap-index.xml or the llms.txt index; robots.txt keeps AI-search crawlers welcome while blocking training scrapers.

Why we did this

The same content is available to humans as plain HTML pages and to AI agents as structured data (see llms.txt). If an assistant cites Onchain Diary when answering a Web3 security question, the URL it references always points back to a readable page — not a raw dump.