convez.ai
MCP ServerBusca & PesquisaIniciante
kecven

SCOUTS-AI MCP Server

kecven/scouts-ai-mcp

Busca web sem API key para agentes Claude e LLMs via MCP

Ver no GitHub
InicialPluginScore · de 0 a 100

Servidor MCP que expõe a API de busca web SCOUTS-AI como uma ferramenta única e sem autenticação para agentes de IA. Integra-se diretamente com Claude Desktop, Cursor, Continue e outros hosts MCP, com suporte a stdio e HTTP.

Instalação

$npx convez-ai kecven/scouts-ai-mcp

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Agentes de IA que precisam buscar informações na web em tempo real
  • Answer engines e workflows de pesquisa com contexto atualizado
  • Integração em aplicações Claude Desktop, Cursor e Open WebUI
  • Self-hosted MCP bridges com transporte HTTP ou Docker

Veredito

Ferramenta prática e leve que democratiza o acesso à busca web em agentes IA, eliminando overhead de API keys. Implementação limpa e bem documentada com suporte multi-plataforma (stdio, HTTP, Docker), ideal para quem quer busca contextual sem complexidade.

README

A tradução para o português deste README ainda está na fila.

scouts-ai-mcp

Model Context Protocol (MCP) server that exposes the SCOUTS-AI web search API as a single web_search tool for AI agents, LLM apps, answer engines and GEO workflows.

  • One tool, no API key. Backed by GET https://scouts-ai.com/api/search.
  • Drop-in for Claude Desktop, Cursor, Open WebUI, Continue, Cline and any MCP host.
  • Python ≥ 3.10, fastmcp v2, httpx.
  • MIT licensed.

Install

pip install scouts-ai-mcp

Run (stdio)

scouts-ai-mcp

That's it. Wire it into your MCP host of choice — for example, Claude Desktop's claude_desktop_config.json:

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

Run (HTTP)

For remote MCP hosts and self-hosted bridges:

scouts-ai-mcp --transport http --host 127.0.0.1 --port 8765

Run (Docker)

A multi-arch (linux/amd64 + linux/arm64) image is published to Docker Hub as kecven/scouts-ai-mcp. It runs the MCP server in streamable HTTP mode on port 8765 and exposes the web_search tool at http://localhost:8765/mcp. No API key, no external dependencies.

Quick start:

docker run --rm -p 8765:8765 kecven/scouts-ai-mcp:0.1.6

Then point any streamable-HTTP MCP host at http://localhost:8765/mcp (or, when run behind a public proxy, https://<your-host>/mcp).

Override the upstream API base URL:

docker run --rm -p 8765:8765 \
  -e SCOUTS_AI_BASE_URL=https://scouts-ai.com \
  kecven/scouts-ai-mcp:0.1.6

Append CLI args (the image entrypoint is scouts-ai-mcp):

docker run --rm -p 8765:8765 kecven/scouts-ai-mcp:0.1.6 --log-level=DEBUG

Available tags:

  • kecven/scouts-ai-mcp:0.1.6 — pinned, recommended for production.
  • kecven/scouts-ai-mcp:0.1 — minor-version rolling tag.
  • kecven/scouts-ai-mcp:latest — latest stable release.

Build and push locally (requires docker buildx):

docker buildx build \
  --platform linux/amd64,linux/arm64 \
  -f Dockerfile.hosted \
  -t kecven/scouts-ai-mcp:0.1.6 \
  -t kecven/scouts-ai-mcp:0.1 \
  -t kecven/scouts-ai-mcp:latest \
  --push .

The image runs as a non-root user (scouts, uid 1001) and includes a TCP-level HEALTHCHECK on 127.0.0.1:8765 (the streamable HTTP endpoint does not return 200 on plain GET, so an HTTP probe would be unreliable).

Tool: web_search

ParameterTypeDefaultDescription
querystring—Search query, 1–512 chars.
langstringenBCP-47 language code (e.g. en, en-US).
pageint11-based page number, 1–10.

Returns a compact JSON object mirroring the SCOUTS-AI response shape:

{
  "query": "rust async runtime",
  "lang": "en",
  "page": 1,
  "pageSize": 10,
  "cached": false,
  "tookMs": 412,
  "results": [
    {
      "title": "Tokio - An asynchronous runtime for Rust",
      "url": "https://tokio.rs/",
      "content": "Tokio is an asynchronous runtime for the Rust programming language...",
      "publishedAt": "2025-11-14T00:00:00Z",
      "engine": "duckduckgo"
    }
  ]
}

Error handling

The tool raises ToolError (rendered as an MCP tool error) when:

  • The query is empty/too long or lang/page are invalid → invalid arguments.
  • The upstream returns 429 → rate limit exceeded; honors Retry-After when present.
  • The upstream returns 5xx or the network call fails → SCOUTS-AI temporarily unavailable.
  • The upstream returns a structured 4xx error envelope → forwards the code and message.

Configuration

All settings are environment variables. Defaults match the public SCOUTS-AI deployment.

VariableDefaultDescription
SCOUTS_AI_BASE_URLhttps://scouts-ai.comBase URL of the SCOUTS-AI API.
SCOUTS_AI_TIMEOUT_S5.0HTTP timeout in seconds (0.1–60).
SCOUTS_AI_USER_AGENTscouts-ai-mcp/0.1.6User-Agent header.
SCOUTS_AI_DEFAULT_LANGenDefault lang when the tool omits it.
SCOUTS_AI_MAX_QUERY_LENGTH512Reject queries longer than this.
SCOUTS_AI_MAX_PAGE10Reject page numbers above this.

Development

git clone https://github.com/kecven/scouts-ai-mcp.git
cd scouts-ai-mcp
python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
pytest

License

MIT — see LICENSE.

Ver o original no GitHub

Bom para

Relacionados