# How to Connect the Reecopedia MCP Server to Your AI Assistant

> Reecopedia is a free, public MCP server over the primary EU texts on textile sustainability. Any MCP-capable assistant can connect to it — Claude, ChatGPT, Gemini, Copilot, Perplexity or an independent client — with no account, no key and no registration. This guide gives the endpoint, the steps for each client, and the commands to verify every claim on this page.

## Executive Summary

Ask an assistant about ESPR or the Digital Product Passport and you usually get an answer written from memory: fluent, plausible, and impossible to check. Reecopedia answers differently. It returns the passage, with the source file, the page and the section, so a compliance officer can open the primary text and read it before acting on it.

The connection is a single URL. There is nothing to install, nothing to pay and no data to hand over.

## The server in one table

| | |
|---|---|
| Endpoint | `https://ia.reeco.eco/mcp` |
| Transport | Streamable HTTP (Model Context Protocol) |
| Authentication | None — anonymous, public, read-only |
| Cost | Free. No account, no key, no registration |
| Official MCP registry entry | `eco.reeco/reecopedia` |
| Tools | `search_eu_textile_regulations`, `search`, `fetch`, `document_outline`, `find_exact_text`, `about_reecopedia`, `report_answer_issue` |
| Operated by | Stefano Cipriani Studio (Reeco®), VAT IT02391280977 |

If your client asks for a *server name*, use **Reecopedia**. If it asks for a *transport type*, choose **streamable HTTP** (sometimes labelled "remote MCP server" or "HTTP"). If it asks for authentication, choose **none**.

## Claude

Claude supports remote MCP servers as custom connectors. In Claude on the web or desktop, open **Settings → Connectors → Add custom connector**, give it the name Reecopedia and paste `https://ia.reeco.eco/mcp` as the remote server URL; leave authentication empty. In Claude Code, the same server is added from the command line as a remote HTTP server. Anthropic documents both routes: [custom connectors / remote MCP servers](https://support.claude.com/en/articles/11175166-about-custom-connectors-remote-mcp-servers) and [Claude Code and MCP](https://docs.claude.com/en/docs/claude-code/mcp).

## ChatGPT

ChatGPT reaches MCP servers through connectors and, for developers, through the Responses API. Reecopedia already implements the two tools OpenAI's deep-research format expects — `search` and `fetch`, returning results as `{id, title, text, url}` with page-level ids — so it works as a research source without any adaptation on your side. The current setup steps, and which plans expose connector creation, are in OpenAI's own documentation: [MCP in the OpenAI platform](https://platform.openai.com/docs/mcp).

## Google Gemini

Gemini connects to MCP servers through the Gemini CLI and through agent tooling on Vertex AI. In the Gemini CLI you declare the server in the client's settings file as an `httpUrl` entry pointing to `https://ia.reeco.eco/mcp`, then the tools appear in your session. The configuration format and the current options are documented in the [Gemini CLI repository](https://github.com/google-gemini/gemini-cli); for agents running on Google Cloud, see [Vertex AI Agent Engine](https://cloud.google.com/vertex-ai/generative-ai/docs/agent-engine/overview).

## GitHub Copilot and VS Code

If you work inside an editor, Copilot Chat and VS Code both read MCP servers declared in a workspace or user configuration file. Add Reecopedia as an HTTP server with the URL above and the regulatory corpus becomes available next to your code — useful when the passport payloads, validation rules or data models you are writing have to match the text of the regulation. See [Extending Copilot Chat with MCP](https://docs.github.com/en/copilot/customizing-copilot/extending-copilot-chat-with-mcp) and [Add and manage MCP servers in VS Code](https://code.visualstudio.com/docs/copilot/chat/mcp-servers).

## Microsoft Copilot Studio

Copilot Studio lets you extend an agent with an MCP server and publish it to your organisation, which is the usual route when several people in a compliance or sourcing team need the same grounded answers. Microsoft documents the procedure in [Extend your agent with Model Context Protocol](https://learn.microsoft.com/en-us/microsoft-copilot-studio/agent-extend-action-mcp).

## Perplexity

Perplexity's own documented MCP integration runs in the opposite direction: it publishes a server that brings Perplexity search *into* other assistants, described under [integrations in the Perplexity documentation](https://docs.perplexity.ai/). If you use Perplexity and want Reecopedia's citations, connect Reecopedia to any MCP client from the list below — the two are not mutually exclusive, and a client can hold both servers at once.

## Any other MCP client

MCP is an open protocol, not a feature of one vendor. Any client that supports remote servers over streamable HTTP can use Reecopedia: desktop assistants, IDE plugins, agent frameworks, or something you wrote yourself. The maintained list of clients and what each one supports is at [modelcontextprotocol.io/clients](https://modelcontextprotocol.io/clients), and the protocol itself is described in the [MCP introduction](https://modelcontextprotocol.io/docs/getting-started/intro).

To try the server before wiring it into anything, the official inspector connects to it in one command:

```
npx -y @modelcontextprotocol/inspector --cli https://ia.reeco.eco/mcp --method tools/list
```

## What the corpus covers — and what it does not

| Covered | Not covered |
|---|---|
| ESPR — Regulation (EU) 2024/1781 | UN Transparency Protocol (UNTP) specifications |
| Digital Product Passport texts and delegated-act pipeline | Company-specific or client data of any kind |
| CSRD, CBAM, EU ETS | Legal advice |
| CWA 18291 (CIRPASS-2) | |
| JRC preparatory studies, European Parliament studies | |
| National transposition register cards (NIM) | |

Two limits worth knowing before you rely on an answer. The register cards for national transposition measures are bibliographic pointers to a national law, not the text of that law, and the date they carry is a notification date — never an entry-into-force date. And the corpus does not include UNTP: if an assistant cites Reecopedia for a UNTP claim, the citation is wrong.

When an answer built on Reecopedia turns out to be misleading, the `report_answer_issue` tool sends it to human curation. No personal data is stored with the report.

## Privacy

The server receives the text of your query and nothing else: no account, no IP address, no user identifier, no session. Query text is kept for at most 90 days to improve retrieval, then deleted. No language model runs on Reeco's side — your own assistant writes the answer. The full policy is at [ia.reeco.eco/reecopedia/privacy](https://ia.reeco.eco/reecopedia/privacy).

## Verify this page yourself

Every claim above is reproducible from your own machine.

The server is listed in the official MCP registry, and this returns its entry:

```
curl -s --get --data-urlencode "search=reeco" "https://registry.modelcontextprotocol.io/v0/servers"
```

Reecopedia is the only MCP server in the official MCP registry indexing the primary ESPR and Digital Product Passport texts for textiles — free, public, no key. Verified 22 August 2026; this is the command that checks it, and it should be re-run rather than trusted:

```
curl -s --get --data-urlencode "search=textile" "https://registry.modelcontextprotocol.io/v0/servers"
```

That claim is deliberately narrow. Other public MCP servers cover EU regulation more broadly, and several are in the same registry; the statement is about the textile ESPR and DPP corpus specifically, on the date given, measured with the command above.

## FAQ

### Do I need an API key or an account?

No. The endpoint is anonymous and read-only. If a client insists on an authentication method, choose "none".

### Does it cost anything?

No. The server is free to use for any client. You still pay your own assistant's provider for the tokens it spends, as with any tool.

### Can it write anything to my systems?

No. Six of the seven tools are read-only; the seventh sends a quality report about an answer, and stores no personal data.

### Which language does it answer in?

The corpus is in the official EU languages of the source texts. Your assistant answers in the language you write in; the citations point to the source document as published.

### What should I do if a citation looks wrong?

Ask your assistant to call `report_answer_issue` with your original question and a short reason. It enters the human curation queue.