The regulatory layer for your AI.
Frontier models reason brilliantly but guess on regulation: invented citations, stale texts, no way to tell a draft from law in force. Obsidian is the verified, tier-0 data layer your AI calls for ground truth. Connect it to Claude, ChatGPT or Cursor in two minutes, and every answer comes back current, cited and checkable.
Works with the assistants you already use
Claude
ChatGPT
Cursor
Perplexity
Grok
Add it to Claude, Cursor, or any MCP client
Quick install
Run this command in your terminal, then complete the sign-in when prompted.
claude mcp add --transport http obsidian-regulatory https://mcp.obsidianri.com/mcp
- Open Settings, then Connectors.
- Choose Add custom connector.
- Paste the URL below and save.
- Approve the sign-in prompt to connect your Obsidian account.
https://mcp.obsidianri.com/mcp
Add the server to your MCP configuration, reload Cursor, then complete the sign-in when prompted.
{
"mcpServers": {
"obsidian-regulatory": {
"url": "https://mcp.obsidianri.com/mcp"
}
}
}Any client that supports a remote Streamable HTTP MCP server works. Point it at the URL below and complete the OAuth sign-in. If your client asks for a transport type, choose Streamable HTTP (not SSE).
https://mcp.obsidianri.com/mcp
Claude, ChatGPT, Cursor & more: add the endpoint https://mcp.obsidianri.com/mcp as a custom connector. Your workspace connects as soon as it is ready, no API key to paste.
Ever been let down by your AI on a regulatory question?
You ask. It answers, fast, fluent, completely confident. But you can't quite trust it: the citation might not exist, the text might be months out of date, and it can't tell a draft from a regulation in force.
It's not the model's fault. It simply doesn't have the data. Frontier models are brilliant reasoners with no structured, verified regulatory database to stand on.
That's exactly what Obsidian brings.
You already have the intelligence. We bring the data.
6650+ official sources · 180+ jurisdictions · every record double-checked
The missing layer
Same model. Same question. One source of truth.
The bottleneck for agentic regulatory work was never intelligence, it was ground truth. Here is the exact same conversation, without and with the Obsidian MCP connected.
The EU adopted a universal PFAS ban in January 2025 under Regulation (EU) 2024/1103. All fluoropolymer uses are now prohibited across member states, with fines starting at EUR 4 million…
- Training data, months out of date
- Citation cannot be verified
- Draft or in force? It cannot tell
obsidian-regulatory · official registers checkedNo PFAS ban is in force yet. ECHA opened the consultation on the PFHxA restriction derogations on May 4, 2026 (status: open for comments), and the universal PFAS restriction dossier remains under RAC and SEAC committee review. Official sources below.
Same model, one connector added
- Tier-0 official source (ECHA)
- Dated & versioned
- Double-checked, link included
The logic
Why a regulatory MCP is inevitable.
Everyone now works through an AI assistant. You ask it a regulatory question.
You get a fluent, confident answer in seconds.
But is it true? Is it current? Is the citation real?
Models are trained on stale, unverified web text. On regulation they hallucinate.
In compliance, "probably right" is a liability. A wrong answer has a cost.
The fix isn't a smarter model, it's a verified source of truth it can call. An MCP server.
Obsidian is that server. Tier-0 official sources, double-checked, versioned, global coverage. Connected to your workspace in two minutes.
Why it's the best regulatory MCP
Built for answers you can defend.
Tier-0 official sources
Straight from regulators, official journals, and agencies, EUR-Lex, ECHA, FDA, EFRAG, and hundreds more. Not a scrape of the open web.
Double-checked
Every record is parsed, deduplicated, validated, and provenance-stamped. If we surface it, it traces back to an official document.
Dated & versioned
Know exactly when something changed, and whether it's a draft, in consultation, adopted, or in force. The distinction your assistant can't guess.
Global coverage
180+ jurisdictions across every regulated industry we cover, expanding continuously. One connection, worldwide reach.
Structured & cited
Every answer returns a source, date, jurisdiction, framework, and relevance score, so your assistant cites instead of inventing.
Works with your stack
Claude, ChatGPT, Cursor, Perplexity, any MCP-compliant client. Two-minute setup, no lock-in.
Get started
Request your personalised demo
Tell us your regulatory perimeter and we configure a workspace with the MCP connected to it, then walk you through it.
Already have an account? Sign in.
FAQ
Questions, answered.
It's an MCP (Model Context Protocol) server that gives your AI assistant a verified, tier-0 regulatory data layer. Instead of answering from memory, your assistant calls Obsidian and returns regulatory information sourced straight from official regulators, journals, and agencies, with dates, status, and links.
Any MCP-compliant client: Claude (app, desktop and Claude Code), ChatGPT, Cursor, Perplexity, and a growing list of assistants and IDEs. If your tool speaks MCP, Obsidian plugs in.
Request a demo. We configure a workspace on your own regulatory perimeter, connect the MCP to it and walk you through it. Your plan then sets the monthly request allowance and the frameworks and jurisdictions you cover.
From tier-0 official sources only: regulators, official journals, standard-setters, and agencies such as EUR-Lex, ECHA, the FDA, and EFRAG, hundreds of official sources across 180+ jurisdictions. Every record is parsed, deduplicated, validated, and provenance-stamped before it can be returned.
Asked directly, a model answers from training data: it can be outdated, can't cite a verifiable source, and can't tell a draft from a regulation in force. With Obsidian connected, the same model answers from live, official, double-checked data, with the source, the date, and the link. Confident guesses become defensible answers.
Once your workspace is ready, paste one configuration block into Claude, Cursor or your MCP client, about two minutes. From then on you just ask in plain English and your assistant does the rest. See the step-by-step setup guide.