MCP
MCP overview
The Obsidian MCP server gives any MCP‑capable AI assistant a verified regulatory layer. Your assistant keeps its own reasoning — Obsidian supplies the official, sourced facts it answers from.
What is MCP?
The Model Context Protocol is an open standard for connecting AI assistants to external tools and data. When you add the Obsidian MCP server to a client like Claude, ChatGPT or Cursor, the assistant gains a regulatory‑search tool it can call on its own, whenever a question touches regulation.
Endpoint & transport
The server speaks the Streamable HTTP transport at a single endpoint:
https://mcp.obsidianri.com/mcp
The legacy /sse transport has been retired. Use the /mcp Streamable HTTP endpoint above.
Authentication
Access is tied to your Obsidian account through OAuth. The first time your client connects, it opens a one‑click sign‑in that links the MCP session to your account — no API key to copy or paste. See Connect your client for the per‑client flow.
Available tools
The server currently exposes one primary tool:
| Tool | Purpose |
|---|---|
regulatory_search | Search verified, tier‑0 official regulations by meaning — obligations, deadlines, legal status, and what changed — filtered by jurisdiction, framework and status. Full reference in Tools & reference. |
How results are scoped
Results are automatically limited to the industries your account covers (ESG & sustainable finance, chemicals & advanced materials, life sciences & medtech). You do not need to pass an industry filter for that — it is applied server‑side from your account perimeter.
Why it beats an ungrounded model
On regulatory questions, a general model guesses from training data of unknown age. With Obsidian connected, the same model answers from the current official text and returns the reference so the user can verify it. See the regulatory AI benchmark for measured accuracy and citation rates.