MCP Server overview

Connect Claude Code, Claude Desktop, or any MCP-compatible client to query your practice data via natural language. Read-only v1.

Updated 2026-04-24 · Edit this page on GitHub

title: "MCP Server overview" description: "Connect Claude Code, Claude Desktop, or any MCP-compatible client to query your practice data via natural language. Read-only v1." order: 1 category: "api" updated_at: "2026-04-24"

Rounds ships a Model Context Protocol (MCP) server so you can ask natural-language questions about your practice's data from Claude Code, Claude Desktop, or any MCP-compatible client.

What you get

Ten read-only tools covering:

  • Analytics — Revenue Overview, YTD Billables, YoY Comparison
  • Patient flow — list appointments, single appointment detail (anonymous P-XXXX codes)
  • Billables — list with status/provider/payer/CPT filters
  • Audit log — search by action prefix, actor, date range
  • Metadata — modules enabled, integrations status, health check

Every response is PHI-free by design — aggregates + anonymous codes only. Every query is rate-limited + audit-logged.

Quick start

  1. Issue a token at /admin/integrations/mcp (owner-only)
  2. Save the plaintext token — shown once, not recoverable
  3. Add to your MCP client config — full setup walkthrough at github.com/a-c-royce/ralthealth/blob/main/dashboard/docs/mcp-integration.md
  4. Ask questions — "What's our YTD billed revenue?" / "Show me the aging report for Smith in Q1" / "List patient flow for today"

Authoritative docs

The operator-facing docs are on GitHub. They cover:

  • Full walkthrough with Claude Code
  • All 10 tools with input schemas
  • 5 example queries
  • Security notes
  • PHI-safeguard design
  • Rate-limit override mechanics
  • Troubleshooting

→ mcp-integration.md on GitHub

We intentionally don't duplicate those here — the operator doc is the authoritative source and we'd rather it drift once than drift twice.

Rate limits

60 requests/hour per token. Override per-tenant at /admin/integrations/mcp → tenant config (up to 10,000/hour). Exceeding returns HTTP 429 with Retry-After header.

Security

  • Tokens are 32-byte base64url strings prefixed rlt_mcp_
  • argon2id hashed on our side; plaintext never stored
  • Scoped to a single practice — can't cross-tenant
  • Revocable one-click in the admin UI
  • Every query writes to mcp_query_log + audit_log

Also available as REST

The same tools are HTTP-accessible at /api/admin/v1/* — see REST API overview.

If you hit a problem

Most MCP integration issues are in the dashboard operator runbook: → runbook.md § MCP server operations

Or email support@ralthealth.com.