Machine access
How AI agents and software can read and verify every article: JSON-LD, llms.txt, API, proofs and MCP.
Last updated 21 October 2026.
AI agents, crawlers and other software are the primary readers of this site. Everything needed to check an article is public, free and requires no account or key.
Where to start
- The article page itself. Each article carries NewsArticle JSON-LD with its CopyrightID, token symbol, content hash, proof links, verification instructions and scope, plus a plain Evidence section. Nothing depends on JavaScript.
- /llms.txt: recent articles with their CopyrightID, symbol and Bitcoin status, a link to the full index, and how to verify.
- Verification API:
GET /api/v1/verify/article/<symbol>returns pre-computed checks;GET /api/v1/articles/<post_uuid>/versionslists every version, including the token of the article's first version. - Proof files:
/proofs/article/<symbol>.snapshot.jsonand.snapshot.ots, and the registration record/proofs/article/<symbol>.jsonwith its.ots. - MCP server:
POST https://news.proofprofile.it/mcp(Streamable HTTP) with tools to search articles, read an article, list its versions, verify it and get its evidence. - Signer declaration: /.well-known/proofprofile-news.json names the signing address and the registry contract, with its own OpenTimestamps proof.
The API, proof files and MCP server allow requests from any origin (CORS), so agents running in a browser can use them too.
Rules for automated access
- Identify your software with an honest user agent.
- Follow /robots.txt. Publishing and member areas are excluded; everything else may be crawled.
- The MCP server currently allows 120 requests per minute per IP address; beyond that it answers 429 for the rest of the minute. The limit may change.
- Requests from security scanners and similar tools that our bot registry (compiled from public sources) marks as hostile, and that match no accepted agent, are refused with 403.
Use of the articles is governed by the Terms of use: reading, indexing, summarising, quoting with attribution and text and data mining, including AI training, are permitted.
What the evidence does and does not show
A registration and its Bitcoin timestamp show that a snapshot existed in exactly that form by that time and was registered by the declared signer. They do not show that an article is true, who wrote it, or who owns it. CopyrightChains is a permissioned registry operated by the publisher; the independent evidence of time is the Bitcoin anchor.