MCP Guide
Point Claude, Cursor or any MCP client at wordradar
Endpoint
POST https://YOUR_DOMAIN/mcp # streamable HTTP (JSON-RPC 2.0)
Authorization: Bearer wr_... # your API key (account → API Keys)Semantics: GET → 405 (no SSE), notifications → 202, DELETE → 204, batch → rejected.
Client config
Claude Desktop / Cursor / any MCP client:
{
"mcpServers": {
"wordradar": {
"type": "http",
"url": "https://YOUR_DOMAIN/mcp",
"headers": { "Authorization": "Bearer wr_YOUR_KEY" }
}
}
}Tools
| Tool | Credits | Args |
|---|---|---|
ledger_pulse | free | — |
ledger_words | 1 | type(新词|二次爆火|长期大词|存量), direction(AI|游戏|工具), site(站群匹配), limit |
proxy_kd_score | 1 | keyword?, results: [{url, domain?}] (top-10 SERP) |
site_radar_latest | 1 | hosting(free|custom|all), min_points, limit |
trends_rising | 2 | keyword |
domain_timeline | 2 | domain |
stripe_leaderboard | 1 | month? (YYYYMM) |
get_usage | free | — |
Billing discipline: a tool is charged only after it succeeds — errors and invalid input are free.
Example: agent-driven niche hunt
Ask your agent:
"Use ledger_words to pull new AI-direction words from the latest period, score the top 5 SERPs with proxy_kd_score, then check which sites already exist for the winners via domain_timeline."
The agent chains the three tools with your key; every call is metered per the table above.