Shared contract

• Endpoint: GET https://scouts-ai.com/api/search
• Params: q (required, 1–512 chars), lang (optional, BCP-47-like or all; omit to send no language filter), page (optional, 1–10, default 1).
• Top-level JSON: query, lang, page, pageSize, cached, tookMs, results.
• Each item in results: title, url, content, publishedAt (or null), engine (or null).
• MCP tool name: web_search.
• OpenClaw provider id: scouts-ai-search. Tool name: scouts_ai_web_search.
• OpenClaw skill (ClawHub slug scouts-ai-skill): teaches any OpenClaw host to call GET /api/search via curl. No plugin, no API key.
• Errors: 429 with Retry-After. Do not bypass rate limits.

1. Claude Desktop (MCP)

Add the MCP server to claude_desktop_config.json:

{
  "mcpServers": {
    "scouts-ai": {
      "command": "scouts-ai-mcp"
    }
  }
}

Test prompt: “Search the web for the latest MCP server registry changes and cite sources.”

2. Cursor (MCP)

Open Cursor → Settings → MCP and add a new global MCP server. Command is the same as Claude Desktop.

{
  "mcpServers": {
    "scouts-ai": {
      "command": "scouts-ai-mcp"
    }
  }
}

Use it for code research, library lookups and changelog checks. Test prompt: “Look up the current TypeScript ESM package.json contract for MCP plugins.”

3. Cline (MCP)

Cline reads the same MCP config. Point its MCP server list to:

{
  "scouts-ai": {
    "command": "scouts-ai-mcp"
  }
}

Test prompt: “Search docs for InetAddress.ofLiteral in Java 25.”

4. Continue (MCP or HTTP)

If your Continue build supports MCP, register the same server. Otherwise point it at the HTTP endpoint:

GET https://scouts-ai.com/api/search?q={query}&lang=en&page=1

5. Open WebUI (MCP)

Open WebUI accepts MCP servers under Settings → Tools. Set command to scouts-ai-mcp and pick a stable name like scouts-ai.

{
  "scouts-ai": {
    "command": "scouts-ai-mcp"
  }
}

Pair it with a local model for private agent search. The tool shows up in chat as web_search.

6. OpenClaw (plugin)

Install the OpenClaw plugin:

openclaw plugins install clawhub:@scouts-ai/openclaw-search

Or from a local checkout:

git clone https://github.com/scouts-ai/scouts-ai-openclaw-plugin.git
cd scouts-ai-openclaw-plugin
pnpm install
pnpm build
openclaw plugins install --link .

Optional config in ~/.openclaw/openclaw.json:

{
  "plugins": {
    "entries": {
      "scouts-ai-search": {
        "enabled": true,
        "config": {
          "baseUrl": "https://scouts-ai.com",
          "defaultLang": "en",
          "timeoutMs": 5000,
          "maxQueryLength": 512,
          "maxPage": 10
        }
      }
    }
  }
}

Tool: scouts_ai_web_search. Enable in tools.allow if your host filters tools by name.

6b. OpenClaw (skill)

Prefer zero-runtime, zero-install setup? Use the SCOUTS-AI Search skill on ClawHub. The skill is a single SKILL.md that teaches any OpenClaw host to call GET /api/search with curl — no plugin, no API key, no MCP server. The host's exec tool runs the request, the skill enforces status checks, temp-dir cleanup (umask 077 + trap … rm -rf) and 429 / 5xx / 4xx handling.

openclaw skills install scouts-ai-skill

Or from a local checkout (no ClawHub account needed):

git clone https://github.com/scouts-ai/scouts-ai-openclaw-skill.git
cd scouts-ai-openclaw-skill
openclaw skills install . --as scouts-ai-search

Start a new session so the agent sees the refreshed skill list (/new). The skill exposes the same contract as the plugin: GET https://scouts-ai.com/api/search?q=…&lang=…&page=…, returns the same JSON shape, and respects Retry-After on 429.

7. Hermes (MCP or HTTP wrapper)

Hermes agents that support MCP can register the official server:

{
  "mcpServers": {
    "scouts-ai": {
      "command": "scouts-ai-mcp"
    }
  }
}

Hermes builds without MCP support can wrap the HTTP endpoint. Example wrapper exposing a web_search tool to a Hermes agent runtime:

import httpx

def web_search(query: str, lang: str = "en", page: int = 1) -> dict:
    """Search the public web via SCOUTS-AI and return compact JSON."""
    r = httpx.get(
        "https://scouts-ai.com/api/search",
        params={"q": query, "lang": lang, "page": page},
        timeout=10.0,
    )
    r.raise_for_status()
    return r.json()

Hook the wrapper into your Hermes tool registry under the name web_search. The exact registration call depends on your Hermes build — check your host’s tool/plugin docs.

8. LangChain (Python tool)

import requests
from langchain_core.tools import tool

@tool
def scouts_ai_web_search(query: str, lang: str = "en", page: int = 1) -> dict:
    """Search the public web via SCOUTS-AI and return compact JSON."""
    r = requests.get(
        "https://scouts-ai.com/api/search",
        params={"q": query, "lang": lang, "page": page},
        timeout=10,
    )
    r.raise_for_status()
    return r.json()

9. LlamaIndex (tool)

import requests
from llama_index.core.tools import FunctionTool

def scouts_ai_web_search(query: str, lang: str = "en", page: int = 1) -> dict:
    r = requests.get(
        "https://scouts-ai.com/api/search",
        params={"q": query, "lang": lang, "page": page},
        timeout=10,
    )
    r.raise_for_status()
    return r.json()

tool = FunctionTool.from_defaults(
    fn=scouts_ai_web_search,
    name="scouts_ai_web_search",
    description="Search the public web via SCOUTS-AI.",
)

10. Dify (HTTP tool)

In Dify, create a new custom tool and add an HTTP request step with this configuration:

Method:  GET
URL:     https://scouts-ai.com/api/search
Headers: (none required)
Query:
  q:    {{ input.query }}
  lang: {{ input.lang | default("en") }}
  page: {{ input.page | default(1) }}

Output variable: json. Map {{ json.results }} to your workflow as a list of {title, url, content}.

11. Flowise (Custom tool node)

In a Flowise agent flow, drop a Custom Tool node and paste:

import requests

def scouts_ai_web_search(query: str, lang: str = "en", page: int = 1) -> dict:
    r = requests.get(
        "https://scouts-ai.com/api/search",
        params={"q": query, "lang": lang, "page": page},
        timeout=10,
    )
    r.raise_for_status()
    return r.json()

Name the tool scouts_ai_web_search and describe it as “Search the public web via SCOUTS-AI.”

12. n8n (HTTP Request node)

Add an HTTP Request node to your n8n workflow:

Method:        GET
URL:           https://scouts-ai.com/api/search
Authentication: None
Query Parameters:
  Name: q     Value: {{ $json.query }}
  Name: lang  Value: {{ $json.lang ?? "en" }}
  Name: page  Value: {{ $json.page ?? 1 }}
Response:      JSON
Options → Timeout: 10000

Use the returned results array in the next n8n node (Agent, Set, Email, Slack, etc.).

Troubleshooting

• 429 RATE_LIMIT_EXCEEDED: read Retry-After and back off. Do not rotate IPs.
• 503 UPSTREAM_UNAVAILABLE: tell the user search is temporarily unavailable and retry later.
• BAD_REQUEST: q is blank or longer than 512 chars, page is out of 1–10, or lang is not a valid BCP-47-like tag.
• MCP server not visible in host: restart the host so it re-reads the MCP config, or check the host’s MCP config file path.
• OpenClaw plugin not loading: confirm openclaw is the host’s plugin runtime and that scouts-ai-search is enabled.