AI-native

The tea API as an MCP server.

Connect Claude Desktop, Cowork, or any MCP client to the Chinese tea encyclopedia — teas and topic articles in 72 languages with brewing, terroir, chemistry, production steps, verified geographical indications and a multilingual glossary — and to the b2btea.com directory of tea companies. The tools are listed below; live counts are in llms.txt.

Hosted — connect by URL no install

Point any MCP client that supports remote (Streamable HTTP) servers at:

https://api.thetea.app/mcp

Add an Authorization: Bearer tt_… header to use a paid key — free works without one (English only).

Clients that read remote servers from JSON:

{
  "mcpServers": {
    "thetea": {
      "url": "https://api.thetea.app/mcp",
      "headers": { "Authorization": "Bearer tt_your_key" }
    }
  }
}

Stdio-only clients (some Claude Desktop builds) can bridge with mcp-remote:

{
  "mcpServers": {
    "thetea": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://api.thetea.app/mcp",
               "--header", "Authorization: Bearer tt_your_key"]
    }
  }
}

Alternative — local Python server

1. Install the runtime: pip install "mcp[cli]" httpx
2. Save server.py somewhere, note its absolute path.
3. Add it to claude_desktop_config.json:

{
  "mcpServers": {
    "thetea": {
      "command": "python3",
      "args": ["/absolute/path/to/server.py"],
      "env": { "THETEA_KEY": "tt_your_paid_key" }
    }
  }
}

A paid key from the pricing page unlocks 7 languages (Pro) or all 72 (Expert / Lifetime).

Tools exposed

Tea encyclopedia

• search_tea
• get_tea
• get_tea_card
• preview_tea
• get_field
• list_teas
• compare_teas
• random_tea
• list_topics
• similar_teas
• semantic_search
• recommend_like
• ask_sommelier
• tea_map

Reference

• glossary
• get_term
• find_places
• tea_families
• catalog_meta
• list_infusions
• get_infusion

Company directory

• search_companies
• get_company
• company_facets
• companies_for_tea

get_tea returns Markdown for reading; get_tea_card returns the same card as JSON with every field. catalog_meta lists the section and field codes that get_field accepts. Summaries written by a model are marked as machine drafts.

Company directory

The same server opens b2btea.com, a directory of tea companies worldwide — factories, wholesalers, importers, exporters, tea houses, markets and cooperatives — in 15 languages: English, Russian, Chinese, German, French, Spanish, Italian, Portuguese, Japanese, Korean, Turkish, Arabic, Polish, Dutch and Vietnamese.

• search_companies filters by country, city, Chinese province, company type, tea, Chinese-tea focus and evidence level; each result links to its page on b2btea.com.
• get_company returns the card as published on b2btea.com: description, address, phone, website, contact form, teas linked to encyclopedia articles.
• company_facets counts companies by country, type, tea and region; companies_for_tea goes from an encyclopedia article to the companies that deal in that tea. The directory tags companies by tea class and a few named teas, so for most articles it falls back to the class and says so in matched.

What you will not find there: e-mail addresses, in any field. Entries come from open sources and do not claim that a company operates today; each carries its evidence level, from a website that answered a live check to a name a source mentioned. Without a key a search returns the first 10 results and no paging; company cards are always open. REST: /api/v2/companies, /api/v2/company/{slug}, /api/v2/companies/facets.

Other ways in

The same data is also a plain REST API, an OpenAPI spec for GPT Actions, and an llms.txt. Your plan and usage: Account.