Clarity MCP is a condition-aware ingredient and product safety MCP server designed for health, fitness, tracking, and wellness AI workflows. It provides 9 tools, and every answer includes a conclusion, evidence level (gold/silver/bronze), and citations (if any). The server is hosted at mcp.healthai.com, can be used anonymously without installation or an API key (with rate limits), and has been published to the official MCP registry under the identifier com.healthai/clarity (domain-verified namespace).
Main capabilities:
- Condition-aware checks: supports multiple condition lenses, including breastfeeding, lactation, pregnancy, histamine, mcas, rosacea, hs, allergy, all.
- Toolset: provides 9 tools covering ingredient checks, product safety scans, supplement stack audits, health claim fact-checking, FDA recall lookups, and more.
- Evidence and citations: each result includes a conclusion, evidence level, and citations (if any), and provides a human_url link to the Clarity page as the source.
- MCP prompts and resources: the server also exposes MCP prompts (audit_supplement_stack, fact_check_health_claim, product_safety_check) and resources (clarity://methodology, clarity://lenses).
Usage:
- Hosted endpoint:https://mcp.healthai.com（Streamable HTTP, JSON-RPC 2.0).
- For Claude Desktop, Cursor, or any client that supports native remote MCP, add it to the configured mcpServers.
- For older clients that do not support native remote, use the mcp-remote bridge: "command": "npx", "args": ["-y", "mcp-remote", "https://mcp.healthai.com"]。
- OpenAI Responses API: pass the hosted endpoint as a remote MCP tool, and keep approval enabled during evaluation.
- Anonymous usage is limited to 60 tool calls per IP per minute; adding an X-API-Key header allows for higher metered limits.
- Example code: the examples/mcp directory provides examples for direct execution, LangChain, LlamaIndex, Cloudflare Agents, and Hugging Face clients.
Applicable scenarios:
- Ingredient and product safety queries in health, fitness, tracking, and wellness AI workflows.
- Supplement stack audits, health claim fact-checking, product safety scans, and FDA recall lookups.
- Safety checks that require condition awareness (such as pregnancy, breastfeeding, allergies, etc.).
Limitations and notes:
- Anonymous usage has rate limits (60/minute/IP).
- Inputs must not contain personal information (names, contact details, medical records, etc.).
- Results are for reference only and do not constitute medical advice; not being flagged does not mean it is safe.
- Tool calls write bounded operational telemetry (tool name, key/anonymous status, condition lens, request fingerprint, status, duration), and do not include raw parameters.
- scan_barcode may query Open Food Facts or Open Beauty Facts and refresh a private mirror; uncovered validate_claim text may enter a private review queue.
- MCP transport security: non-browser clients may omit Origin; requests that include Origin must exactly match the endpoint origin, otherwise HTTP 403 is returned.
- This repository is the public developer interface for the hosted server, including setup, examples, metadata, and a stdio bridge; the production implementation, deployment configuration, credentials, and operational data are maintained separately.
