Ancient Echoes

Ancient Echoes MCP

Ancient Echoes exposes deterministic Chinese metaphysics computation as a remote MCP server: BaZi (Four Pillars), Zi Wei Dou Shu, Qi Men Dun Jia, Qi Zheng Si Yu, Vedic and Western astrology, Liu Yao, Mei Hua, Da Liu Ren, Xiao Liu Ren, Tai Yi, almanac and election-day evidence.

The computation engine stays on our servers. This repository only contains the client-side skill and setup notes, so any client that speaks MCP can call the same versioned tools.

What you get

Every tool response carries calculationCapability / calculationVersion so results stay reproducible across releases.

Listed in the official MCP Registry as io.github.FlamesCN/ancient-echoes-mcp. Hosts that support MCP Apps render BaZi results as an interactive chart; other hosts get the same structured JSON.

OAuth-capable connectors can discover the authorization server at https://www.ancientechoes.ai/.well-known/oauth-protected-resource and connect with a standard PKCE flow. Developers can keep using a Web API key in the Authorization header.

1. Get an API key

Sign in at https://www.ancientechoes.ai and open User Center → LingXi Devices:

• Web API Key (sk-…): simplest for MCP clients.
• Device token (lxd_…): pair a machine with a short code; it carries the mcp scope by default and can be rotated or revoked without touching your password.

Either credential reaches POST /api/v1/mcp and POST /api/v1/tools/:name.

2. Connect an MCP client

Endpoint: https://www.ancientechoes.ai/api/v1/mcp (Streamable HTTP). The server speaks protocol versions 2026-07-28, 2025-11-25, 2025-06-18 and 2025-03-26.

{
  "mcpServers": {
    "ancient-echoes": {
      "url": "https://www.ancientechoes.ai/api/v1/mcp",
      "transport": "streamable-http",
      "headers": {
        "Authorization": "Bearer sk-your-key"
      }
    }
  }
}

curl -sS https://www.ancientechoes.ai/api/v1/mcp \
  -H "Authorization: Bearer $ANCIENT_ECHOES_TOKEN" \
  -H "Content-Type: application/json" \
  -H "MCP-Protocol-Version: 2026-07-28" \
  -H "Mcp-Method: server/discover" \
  -d '{"jsonrpc":"2.0","id":1,"method":"server/discover","params":{"_meta":{"io.modelcontextprotocol/protocolVersion":"2026-07-28"}}}'

3. Call a tool

curl -sS https://www.ancientechoes.ai/api/v1/tools/bazi.calculate_chart \
  -H "Authorization: Bearer $ANCIENT_ECHOES_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"input_mode":"birth","year":1990,"month":6,"day":15,"hour":9,"minute":30,
       "is_male":true,"calendar_type":"solar","is_leap_month":false,
       "cast_time_source":"birth","timezone_id":"Asia/Shanghai",
       "split_zi_hour":true,"use_true_time":false,"apply_dst":false}'

Rules that matter:

• Pass an explicit civil time, calendar, timezone and (for true solar time) coordinates. The server never fills missing fields with its own clock or a default city.
• Paid tools return tool_entitlement_required when your plan does not include the feature.
• Quota: 200 tool calls/day and 30 calls/minute per user by default (429 + tool_quota_exceeded when exceeded).
• All current tools are read-only. Results are evidence for human reading, not medical, legal, investment or life-decision advice.

Self-hosted stdio bridge (Docker)

Clients and registries that only speak MCP over stdio can run this repository's bridge. It contains no computation engine: initialize, ping and tools/list are answered locally from the bundled catalog (38 tools, generated from the public OpenAPI document), and tools/call is forwarded to POST /api/v1/tools/<name> with your own credential.

docker build -t ancient-echoes-mcp .
printf '%s\n' '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' \
  | docker run --rm -i ancient-echoes-mcp

Set ANCIENT_ECHOES_TOKEN (Web API Key or device token) to enable tools/call; without it the bridge answers discovery requests and returns an explicit isError result for execution. ANCIENT_ECHOES_BASE_URL and ANCIENT_ECHOES_TIMEOUT_MS override the endpoint and timeout.

Skill

ancient-echoes-mcp/SKILL.md teaches an agent how to discover tools, collect the minimum required parameters, respect entitlements, and report results without inventing predictions. Drop it into your agent's skills directory.

License

MIT. The server-side engine, knowledge base and datasets are not part of this repository.