CuRE Clarion · Model Context Protocol
A public, read-only MCP connector for CuRE's methodology and compliance corpus.
CuRE Clarion MCP is a remote Model Context Protocol server that gives an AI assistant like Claude read-only access to CuRE's public methodology catalog, and two advisory linting tools you can run against your own artifacts — a regulatory conformance scan and a TMF Reference Model classifier. It carries no connection to any customer database and never receives or returns patient data.
Clarion is the audited AI gateway at the center of the CuRE platform (see CuRE Clarion). Alongside its tenant-facing gateway, Clarion runs a separate, network-isolated public runtime built specifically for MCP-native AI workbenches: a "lighthouse" surface that exposes CuRE's public methodology, conformance-rule, and TMF-reference corpus to any MCP client, with no authentication required to reach it from Claude. The tools are advisory only — useful inputs to your own workflow, not a submission-grade or 21 CFR Part 11 record.
Tools
Three read-only advisory tools
Every tool is pure over the public corpus that ships with the connector plus whatever artifact you supply on the call — there is no path to tenant or patient data.
methodology_catalog Lists CuRE's public methodology catalog — quality monitoring, pharmacovigilance signal detection, randomization, privacy/re-identification risk, and OMOP characterization — with the open-source statistical oracle each method is validated against. Optionally filter by category or a free-text query.
Example prompt
"Using the CuRE Clarion connector, list the randomization methodologies in the methodology catalog."
conformance_scan Lints a rule set you supply — EMA Annex 11, US FDA 21 CFR Part 11, or CDISC CORE-style dataset conformance — against an artifact you provide (your own system descriptor or dataset metadata) and returns a pass/fail/warn report per rule. Nothing is sent to CuRE data; the artifact is yours.
Example prompt
"Run a 21 CFR Part 11 conformance_scan on this system descriptor: {…}"
tmf_classify Classifies document metadata you supply — filename, title, and document type — into the DIA TMF Reference Model 3.x zone, section, and artifact. Document content is never required, and no CuRE data is read.
Example prompt
"Classify this document into the TMF Reference Model: filename 'signed-protocol-v3.pdf'."
Connect
Add the connector in Claude
- 1
In Claude, open Settings → Connectors and choose to add a custom connector.
- 2
Enter the server URL:
https://mcp.cure.principia.health/mcp - 3
No account, API key, or sign-in is required to use the connector from Claude. Once the connector is listed in Anthropic's Connectors Directory, you'll also be able to add it directly from there instead of entering the URL by hand.
Building a direct integration outside Claude, or need a dedicated lane with its own rate limit? Principia issues OAuth 2.0 client-credentials for direct and partner integrations — contact us to request one.
Data & privacy
No tenant database connection, no patient data
The public-advisory MCP runtime is a separate, network-isolated deployment from every tenant-facing CuRE system. It is built with no code path to a customer database.
What the connector never touches
The connector has no connection to any customer database and never receives or returns patient data. Its only inputs are the public methodology/conformance/TMF corpus that ships with the connector, plus whatever artifact you supply on a given call (your own system descriptor, dataset metadata, or document metadata) — never a CuRE tenant record. Results are advisory inputs to your own workflow, not a regulated 21 CFR Part 11 record.
What is recorded per call
Every call is written to a hash-chained, tamper-evident audit log. Each entry records:
- The tool name invoked
- A SHA-256 hash of the call’s input arguments — never the input content itself
- The outcome status (success or error code)
- A timestamp
- The caller’s source IP, used for per-caller rate limiting
The content of your call — the artifact or document metadata you supply — is processed transiently to compute the result and is not stored.
This page describes the MCP connector specifically. For how Principia handles information collected elsewhere on this website, see our Privacy Policy. For how CuRE builds compliance and tenant isolation into the platform generally, see the Trust Center.
Support
Questions about the connector
For setup help, bug reports, or questions about what the connector does or doesn't do, reach out and we'll follow up.
See the rest of the CuRE Clarion gateway
The MCP connector is the public, inbound side of Clarion. The same app also mediates every AI call inside the platform through one audited, provider-agnostic gateway.