THC Open Mindfulness MCP
A public, read-only Model Context Protocol server for The Holistic Care's approved public mindfulness resources — interactive games, free guided audio practices, plain-language research summaries, a glossary, and the public PanchaVikas framework overview.
- Production endpoint:
https://mcp.theholisticcare.com/mcp (Streamable HTTP)
- No authentication required. No API key, no OAuth, nothing to sign up for.
- Read-only. No tool in this server creates, modifies, or deletes anything.
- Backed entirely by a public REST API —
api.theholisticcare.com (docs, source). This server never connects to Sanity CMS, a database, or any payment system directly.
Table of contents
Why this exists
The Holistic Care already publishes a public REST API of its mindfulness content. This server is a thin MCP wrapper around that API so any MCP-compatible client (Claude Desktop, Claude Code, custom agents, etc.) can discover and use that content through native tool calls instead of a bespoke HTTP integration — with the same content boundaries the REST API itself already enforces.
Architecture
MCP client (Claude, etc.)
│ Streamable HTTP, JSON-RPC
▼
thc-mindfulness-mcp (this repo — Next.js App Router, deployed to Vercel)
│ plain HTTPS GET, no credentials
▼
api.theholisticcare.com (sibling REST API project)
│
▼
Sanity CMS (never touched directly by this server)
- Runtime: Next.js (App Router) on Vercel, Node.js runtime (not Edge — the MCP SDK needs Node APIs).
- MCP SDK:
@modelcontextprotocol/server v2 (createMcpHandler, McpServer), the modern high-level package. @modelcontextprotocol/sdk v1 is kept only as a dev dependency for lower-level type definitions and future in-process test clients.
- Transport: Streamable HTTP, stateless. Every request builds a brand-new
McpServer instance (src/server.ts) — there is no session store, so nothing about one caller's request can leak into another's.
- Validation: Zod v4 schemas for every tool input (
src/schemas.ts), re-validated server-side regardless of what the SDK itself already checks.
- Content boundary: enforced entirely by which endpoints this server calls on the upstream REST API (
src/lib/apiClient.ts, src/lib/constants.ts). The base URL is a compile-time constant; no tool argument can ever change what host this server talks to.
All seven tools are read-only, take no destructive action (readOnlyHint: true, destructiveHint: false in every tool's annotations), and return a small (default 6, max 20) list of normalized resources rather than a bulk dump.
| Tool | What it does | Backed by |
|---|
search_mindfulness_resources | Free-text search across every public resource family at once (blog, games, practices, research, glossary, PanchaVikas), with optional resource_type/category/audience/skill/age filters. | GET /v1/search, filtered client-side |
get_mindfulness_resource | Fetch one specific resource by its type + identifier (slug for most families, stable id for PanchaVikas). | The matching dedicated detail endpoint, e.g. GET /v1/mindfulness-games/{slug} |
find_mindfulness_games | Find interactive mindfulness games, filterable by age, skill, audience, and max duration. | GET /v1/mindfulness-games or GET /v1/search |
find_guided_practices | Find free guided audio practices from the Stillness Library (never paid tracks). | GET /v1/practices or GET /v1/search |
search_mindfulness_research | Search plain-language research summaries, filterable by topic and study type. | GET /v1/whitepapers or GET /v1/search |
lookup_mindfulness_term | Look up one glossary term by name, with an optional pillar filter. | GET /v1/glossary/{slug}, falling back to search |
find_panchavikas_resources | Find public PanchaVikas resources (framework pathways and the public downloadable guide only — never private curriculum). | GET /v1/panchavikas/resources, filtered client-side |
Every list-returning tool shares the output shape { results: Resource[], count: number }; get_mindfulness_resource and lookup_mindfulness_term return { found: boolean, resource: Resource | null }. See each tool's own file under src/tools/ for its exact Zod input schema and a code comment explaining any filter it cannot apply with full fidelity (documented rather than silently ignored — see the note on /v1/search's narrower parameter set below).
A known, documented limitation: the upstream /v1/search endpoint accepts only q/limit/offset, while /v1/resources and the dedicated per-family endpoints accept richer filters (category, audience, skill, age_min/age_max) but no free-text query. Where a tool needs both free-text search AND a filter /v1/search doesn't support server-side, it over-fetches from /v1/search and applies the filter client-side (see src/lib/filters.ts) — this is called out explicitly in the affected tool's description text so an MCP client (and the person using it) knows the filter is real but applied after the fact, not a guarantee from the upstream API itself.
Resources
Five static, informational MCP resources — none of them call the upstream API, so they're always available even if api.theholisticcare.com is temporarily down:
| URI | Contents |
|---|
thc://about | What this server is, who provides it, and links to the underlying API/source. |
thc://taxonomy | The resource types this server can return, and what each one means. |
thc://usage | Fair-use notes: no auth needed, best-effort availability, prefer source_url/canonical_url, keep result sizes small. |
thc://content-rights | The software-vs-content licensing split (MIT code, all-rights-reserved content) — see Licensing below. |
thc://panchavikas/framework | The public PanchaVikas five-element framework overview, with an explicit statement of what it deliberately excludes (private curriculum). |
There are no MCP prompts in this server, by design — the spec for this project explicitly avoided adding prompts just because the protocol supports them; nothing here needed one.
Using it from an MCP client
Any Streamable-HTTP-capable MCP client can connect directly to the production URL. For a claude_desktop_config.json-style remote MCP entry:
{
"mcpServers": {
"thc-open-mindfulness": {
"url": "https://mcp.theholisticcare.com/mcp"
}
}
}
No headers, tokens, or environment variables are required on the client side.
Local development
The MCP endpoint is then available at http://localhost:3000/mcp, and a health check at http://localhost:3000/health.
No secrets are required for local development. The only environment variable this server reads is THC_API_BASE_URL (see .env.example), and it defaults to the production REST API if unset — set it only to point a local instance at a different API (e.g. a locally running copy of thc-open-mindfulness-api).
npm run typecheck
npm run lint
npm run test
npm run test:contract
Testing with MCP Inspector
MCP Inspector is the reference tool for interactively exercising an MCP server's tools and resources.
npx @modelcontextprotocol/inspector
Then connect it to either:
http://localhost:3000/mcp (a local npm run dev instance), orhttps://mcp.theholisticcare.com/mcp (production)
using the "Streamable HTTP" transport option and no authentication. You should see all 7 tools and all 5 resources listed, and be able to call each one interactively.
Deployment
This project deploys to Vercel as a standard Next.js App Router app.
- Import the GitHub repo into a new Vercel project.
- No environment variables are required for a production deploy (
THC_API_BASE_URL defaults to https://api.theholisticcare.com).
- Point the custom domain
mcp.theholisticcare.com at the Vercel project (Vercel dashboard → Domains), and add the corresponding DNS record at the DNS provider for theholisticcare.com.
- After deploying, verify:
https://mcp.theholisticcare.com/health returns {"status":"ok",...}https://mcp.theholisticcare.com/mcp responds correctly to an MCP Inspector session (see above)npm run smoke:production passes (scripts/smoke-production.mts — exercises the live endpoint end-to-end)
Security
See SECURITY.md for the full write-up. In short: no secrets, no writes, no direct Sanity/database access, SSRF-hardened outbound requests, Host/Origin header validation on the HTTP transport, and a stateless request model. Report a vulnerability to info@theholisticcare.com rather than opening a public issue.
Licensing
See LICENSING.md for the full explanation. Short version: the code in this repository is MIT licensed; the mindfulness content it retrieves and returns (games, practices, research summaries, glossary entries, blog excerpts, PanchaVikas overviews) is Copyright © The Holistic Care, all rights reserved, unless a specific resource says otherwise — it is public to read and reference, not to redistribute wholesale. See also ATTRIBUTION.md for how to credit The Holistic Care when you build on top of this server.
Contributing / issues
This is a small, single-purpose server maintained alongside The Holistic Care's public API and website. Bug reports and pull requests are welcome via GitHub Issues. For anything security-related, see SECURITY.md instead of opening a public issue.