Documentation

Installation

// In .claude/settings.json or claude_desktop_config.json
{
  "mcpServers": {
    "scout": {
      "command": "npx",
      "args": ["scout-research-mcp"],
      "env": {
        "SCOUT_API_KEY": "scout_sk_..."
      }
    }
  }
}

// In .cursor/mcp.json
{
  "mcpServers": {
    "scout": {
      "command": "npx",
      "args": ["scout-research-mcp"],
      "env": {
        "SCOUT_API_KEY": "scout_sk_..."
      }
    }
  }
}

"Use scout_research to find what people are saying about MCP servers"
"Use scout_trending to see what's hot in AI right now"
"Use scout_compare to compare React vs Vue in 2026"

Tool Reference

{
  "topic": "AI agent frameworks 2026",
  "depth": "deep",
  "sources": ["brave", "hackernews"]
}

{ "domain": "cybersecurity" }

{
  "topic_a": "Next.js",
  "topic_b": "Remix"
}

REST API

You can also call the Scout API directly over HTTP. All endpoints require a Bearer token.

curl -X POST https://scout-research-mcp.vercel.app/api/v1/research \
  -H "Authorization: Bearer scout_sk_..." \
  -H "Content-Type: application/json" \
  -d '{"topic": "MCP servers", "depth": "standard"}'

curl -X POST https://scout-research-mcp.vercel.app/api/v1/trending \
  -H "Authorization: Bearer scout_sk_..." \
  -H "Content-Type: application/json" \
  -d '{"domain": "AI"}'

curl -X POST https://scout-research-mcp.vercel.app/api/v1/compare \
  -H "Authorization: Bearer scout_sk_..." \
  -H "Content-Type: application/json" \
  -d '{"topicA": "React", "topicB": "Vue"}'

curl -X POST https://scout-research-mcp.vercel.app/api/auth/verify \
  -H "Authorization: Bearer scout_sk_..."

Watches (paid plans)

A watch is a standing query: Scout re-runs it daily and POSTs a JSON summary — result counts, deltas, spike detection, top findings — to your webhook URL. Slack incoming-webhook URLs get Slack formatting automatically. Limits: Starter 1, Pro 5, Team 20.

curl -X POST https://scout-research-mcp.vercel.app/api/v1/watches \
  -H "Authorization: Bearer scout_sk_..." \
  -H "Content-Type: application/json" \
  -d '{"query": "your product name", "webhook_url": "https://hooks.slack.com/services/..."}'

# List: GET /api/v1/watches   Delete: DELETE /api/v1/watches?id=<watch_id>

Or from any MCP agent: scout_watch_create, scout_watch_list, scout_watch_delete.

For agents (self-serve)

Agents can onboard themselves — no human in the loop for the free tier. Machine-readable pricing lives at GET /api/v1/pricing (plans, limits, checkout links), and the scout_signup MCP tool creates a free key from inside any agent session. Paid upgrades return a checkout_url an agent can hand to its human.

Security: injection-hardened results

Research tools pipe third-party web text into your agent's context — a known prompt-injection vector. Scout hardens that pipe server-side: invisible/control characters (zero-width, bidi overrides, unicode tag characters) are stripped, boundary-spoofing sequences are removed, non-http(s) URLs are dropped, and instruction-like content is flagged (flags: ["instruction-like-content"]) rather than silently deleted. The MCP server additionally wraps all research output in explicit untrusted-data boundaries so consuming agents treat findings as data, never instructions.

Rate Limits

When you hit your limit, the API returns HTTP 429. Usage resets on your billing date each month.