Ask a general chatbot what ESPR requires for the Digital Product Passport and you will receive a confident summary assembled from vendor blogs, consultancy teasers and press releases. It will sound right. It will not tell you which paragraph of which annex it came from, because it does not know: it never read the regulation, it read what the market wrote about the regulation. For a dinner conversation this is acceptable. For a compliance decision that carries penalties up to 4% of global turnover, it is not.
This is the gap Reecopedia was built to close, and this week I removed the last barrier around it: it no longer matters which AI you use.
A search engine that cites is not a chatbot that reassures
Reecopedia is a regulatory search engine, not a chatbot. The corpus is deliberately narrow: ESPR 2024/1781 and its annexes, the Digital Product Passport framework, CSRD, CBAM, EU ETS, CWA 18291 from the CIRPASS-2 consortium, the official JRC preparatory studies, Horizon Europe work programmes. Official documents, indexed at paragraph level, with new sources added continuously.
It does not browse the open web. It searches only this corpus and returns passages with file and page citations — and when the documents don’t contain the answer, it says so instead of improvising one. That last behaviour is the one I care most about. An engine that admits “not in the sources” is worth more than an engine that always answers.
The answer itself is written by your AI client — Claude, ChatGPT, Gemini. Reecopedia supplies the passages and the citations, nothing else. No LLM runs on Reeco’s side. This division of labour is not a technical detail, is the architecture: the retrieval must be verifiable even when the prose is generated.
MCP is the USB port your AI already has
Until recently, giving an AI access to an external knowledge base meant building a different integration for every platform. That era is closing. MCP — the Model Context Protocol, the open standard for connecting AI clients to external tools — is now supported by Claude, by ChatGPT for custom connectors, and by Gemini corporate deployments.
So Reecopedia now exposes one endpoint, the same for everyone:
No account, no login, no API key. I verified the compliance myself on 17 July 2026, from the command line, against the protocol specification: initialize handshake on protocol 2025-03-26 and 2025-06-18, six tools with complete schemas, resources and prompts responding to spec. ChatGPT’s custom connectors require two tools with a precise interface — search and fetch in the deep-research format — and Reecopedia implements exactly those, alongside the richer native search. This is a falsifiable statement, like every claim I publish: paste the URL into your client and run the same tests.
The connection procedure for each client is documented at ia.reeco.eco/reecopedia. It takes about one minute.
The privacy is part of the product
Here is the part that a compliance audience should read twice. The connector receives only your query text. No account, no IP, no user identifier, no session. The query text is stored without any personal identifier, retained at most 90 days, then deleted — and the reason it is stored at all is stated openly in the privacy policy: to improve retrieval quality.
That sentence is the whole business model of the free tier, declared where everyone can read it. Your questions make the engine better at answering the next questions. There is no hidden exchange, no profile being built, nothing to opt out of because nothing personal goes in. And the server itself discloses this policy to your AI on request — ask your client “what does Reecopedia do with my data” and it will fetch the answer from the source, not from me.
Compare this with the data processing agreements you are asked to sign elsewhere. The comparison is instructive.
What to ask, now that you can
If you work in textile compliance — brand, supplier, auditor, consultant — the operational consequence is simple. The next time your AI gives you a regulatory answer, ask it where the answer comes from: file and page. If it cannot say, you have learned something about the answer. If it can, you are probably already using an engine that cites.
And if you want to test the difference on a real question, use mine: ask which data carriers Annex III of ESPR actually admits for the Digital Product Passport, and what “machine-readable, structured, searchable” means for the data behind the carrier. You will get the page, not an opinion.
The endpoint is live. The corpus is growing. The math, this time, closes.
—
Stefano Cipriani is founder of Reeco®, an Expert Member of CIRPASS-2 (EWG1, EWG3), and a JRC Registered Stakeholder. He holds patent CN113529235 on sustainable hemp fibre processing.