MCP server

The Model Context Protocol (MCP) lets an AI assistant call tools on your behalf. EarlyWave's MCP server gives Claude, Cursor, VS Code and other MCP clients the same 18 read-only operations as the REST API: trends across every source, coverage by country, your monitors, feeds, folders, articles and saved items.

Create a key with the mcp scope in EarlyWave › Settings › API & MCP (open it). In the snippets below, replace YOUR_KEY with it.

Client setup#

Claude Code#

claude mcp add --transport http earlywave https://api.earlywave.ai/mcp \
  --header "Authorization: Bearer YOUR_KEY"

Check it with claude mcp list. Add --scope user to use it in every project.

Cursor#

In .cursor/mcp.json (this project) or ~/.cursor/mcp.json (every project):

{
  "mcpServers": {
    "earlywave": {
      "url": "https://api.earlywave.ai/mcp",
      "headers": { "Authorization": "Bearer YOUR_KEY" }
    }
  }
}

VS Code#

In .vscode/mcp.json:

{
  "servers": {
    "earlywave": {
      "type": "http",
      "url": "https://api.earlywave.ai/mcp",
      "headers": { "Authorization": "Bearer YOUR_KEY" }
    }
  }
}

The tools appear in Copilot Chat in agent mode.

Claude Desktop#

Claude Desktop starts local servers, so it reaches EarlyWave through the mcp-remote bridge (Node.js required). In claude_desktop_config.json (Settings › Developer › Edit Config):

{
  "mcpServers": {
    "earlywave": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://api.earlywave.ai/mcp",
        "--header",
        "Authorization: Bearer ${EARLYWAVE_API_KEY}"
      ],
      "env": { "EARLYWAVE_API_KEY": "YOUR_KEY" }
    }
  }
}

Restart Claude Desktop after saving.

Windsurf and other clients#

Any client that speaks Streamable HTTP needs two things: the address https://api.earlywave.ai/mcp and the header Authorization: Bearer YOUR_KEY. A client that only starts local servers can use the mcp-remote bridge shown for Claude Desktop.

Tools#

Each tool is one operation of the REST API, with the same arguments and the same JSON answer, returned as text.

Lists take limit (up to 100) and offset, and answer has_more. Countries are ISO alpha-2 codes (FR). Answers longer than 50,000 characters are cut; ask for a smaller limit.

Example prompts#

• "What are the breakout stories in tech since yesterday?" → list_trends with category=tech, stage=breakout, since_hours=24.
• "Which stories about France are covered abroad?" → list_trends with country=FR and reach=international,global.
• "Who reported the rail strike first, and from which country?" → get_trend_articles with sort=oldest.
• "In which countries and languages is this story covered, and when did it peak?" → get_trend_coverage.
• "Summarise what my monitors found this week, with sources." → list_monitors, then get_monitor_results with days=7.
• "Find French feeds about climate I don't follow yet." → discover_feeds with query=climate, country=FR.
• "What did I save this week?" → list_saved.

The server also sends the assistant a short guide to its tools, so plain questions work: you rarely need to name a tool.

Test it without a client#

The server is stateless: each POST stands alone, with no session to open first. List the tools with curl:

curl https://api.earlywave.ai/mcp \
  -H "Authorization: Bearer $EARLYWAVE_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Call one:

curl https://api.earlywave.ai/mcp \
  -H "Authorization: Bearer $EARLYWAVE_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"list_trends","arguments":{"country":"FR","limit":3}}}'

The tool's JSON answer is in result.content[0].text.

Troubleshooting#